Copyright Warning & Restrictions

The copyright law of the United States (Title 17, United States Code) governs the making of photocopies or other reproductions of copyrighted material. Under certain conditions specified in the law, libraries and archives are authorized to furnish a photocopy or other reproduction. One of these specified conditions is that the photocopy or reproduction is not to be used for any purpose other than private study, scholarship, or research. If a, user makes a request for, or later uses, a photocopy or reproduction for purposes in excess of fair use that user may be liable for copyright infringement, This institution reserves the right to refuse to accept a copying order if, in its judgment, fulfillment of the order would involve violation of copyright law. Please Note: The author retains the copyright while the New Jersey Institute of Technology reserves the right to distribute this thesis or dissertation
Printing note: If you do not wish to print this page, then select Pages from: first page # to: last page # on the print dialog screen

The Van Houten library has removed some of the personal information and all signatures from the approval page and biographical sketches of theses and dissertations in order to protect the identity of NJIT graduates and faculty.

ABSTRACT EMISSIVITY OF PATTERNED SILICON WAFERS IN RAPID THERMAL PROCESSING by Markus Rabus
The influence of patterns on emissivity in silicon wafers in rapid thermal processing systems has been investigated. In this study, two experiments with layered and patterned silicon wafers were conducted. The main difference in the experiments is the way in which the temperature was controlled. The first experiment was performed under Open Loop Intensity Control (COLIC). For COLIC, no feedback from the wafer is returned. It is assumed that supplying a certain power level will lead to the desired temperature. The other experiment used the Closed Loop Intensity Control. In this case, a feedback, in the form of temperature deviation is used to adjust the temperature. By using the Stefan-Boltzmann T 4 -law, a heat balance equation describing the incoming and outgoing heat can be derived. This heat balance equation can be used to calculate the spatial temperature differences due to different emissivity of the various thin film layers of patterned wafers. A mathematical model was developed based on the heat balance equation. The mathematical model was verified with experiments. The model showed good agreement with the experiments.

EMISSIVITY OF PATTERNED SILICON WAFERS IN RAPID THERMAL PROCESSING

by Markus Rebus

A Thesis Submitted to the Faculty of New Jersey Institute of Technology in Partial Fulfillment of the Requirements for the Degree of Master of Science in Applied Physics Federated Physics Department August 2005

APPROVAL PAGE EMISSIVITY OF PATTERNED SILICON WAFERS IN RAPID THERMAL PROCESSING Markus Rebus

Dr. Anthony T. Fiery, Thesis Co-Advisor Research Professor of Physics, NJIT

Date

Dr. Nuggehalli M. Rkvindra, Thesis Co-Advisor Professor of Physics, NJIT

Date

Dr. Dentcho Ivanov, Committee Member Director of Microelectronics Fabrication Center, NJIT

Date

Dr. Sufi M. Abedrabbo, Committee Member Assistant Professor of Physics, University of Jordan

Date

Dr. Laminae M. Dieng, Committee Member Assistant Professor of Physics, City University of New York

bate

BIOGRAPHICAL SKETCH

Author: Degree: Date:

Markus Rebus Master of Science August 2005

Undergraduate and Graduate Education: Master of Science in Physics, New Jersey Institute of Technology, Newark, NJ, 2005 Diploma (FHA) Wirtschaflsingenieurwesen, University of Applied Science, Kempten, Germany, 2003 Applied Physics

Major:

 my beloved fiancee, Eva Garcia DIaz

ACKNOWLEDGMENT

I would like to express my deepest appreciation to Professors Anthony T. Fiery and. Nuggehalli M. Ravindra who not only served as my research supervisors, providing valuable and countless resources, insight, and intuition, but also constantly gave me support, encouragement, and reassurance. Special thanks are given to Doctors Dentcho Ivanov,. Sufi M. Abedrabbo and Laminae M. Dieng for actively participating in my thesis committee. Thanks to Mr. William Mansfield of Lucent Technology for the valuable data used in the thesis. The experiments were performed at the New Jersey Nano Consortium at Lucent Technologies in Murray Hill, New Jersey. Many of my fellow graduate students in the Applied Physics program are deserving of recognition for their support. I thank the Fuibright Organization for their help and financial support. Lastly, I would like to thank my family and my fiancee for their patience and understanding.

vi

TABLE OF CONTENTS Page ....... 1 4 .. 4 11 17

Chapter 1 INTRODUCTION 2 THEORETICAL BACKGROUND 2.1 Radiation, Emissivity and Heat Conduction 2.2 Rapid Thermal Processing System 2.1.1 Furnace System 2.1.2 Lamp System 2.1.3 Temperature measurement equipment 2.3 Multi-Rad Version 1.0 for Windows 95 and NT 2.3.1 Introduction into Multi-Rad 2.3.2 Calculation Method used by Multi-Rad 2.3.3 Adding Materials
.

20 29 33

..

33 34 42 43

2.4 Derivation of the Mathematical Model 3 SPECIFIC PROCESS MODES AND EXPERIMENTS 3.1 Open-Loop Intensity Control (COLIC) 3.2 Closed Loop Temperature Control (CELT) 3.3 Model Validation Experiments 4 RESULTS AND DISCUSSION APPENDIX A DERIVATION OF THE HEAT BALANCE EQUATION ... ...

52 52 58 62 74 79

vii

LIST OF TABLES

Table 2.1 2.2 2.3 3.1 3.2 Overview of the Different Process Modes Overview of Temperature Measurement Equipment Values of Cosines and at High Symmetry Points Wafer Backside Films Used in the Experiment Temperatures Measured with Thermocouple, Pyrometer and Ripple Pyrometer 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 Thermawave Measurements of Oxide Thicknesses Measured Specifications of Layers on Bulk, Epitaphial and SOIL Wafers Measured Specifications of Bulk, Epitaphial and SOIL Wafers. Calculated Emissivity at 1050 C ... ..

Page 28 33 50 53

56 57 66 67 .. 70 70

Calculated Absorptivity for Furnace Process (Front side) at 1250 C ......... Calculated Absorptivity for Furnace Process (Backside) at 850 C .. Calculated Absorptivity for Lamp System at 2377 C Calculated Effective Emissivity for Furnace at 1050 C Calculated Effective Absorptivity for Furnace Process (Front side) at 1250 C ..

.... 71 71 .. 72

72

3.12

Calculated Effective Absorptivity for Furnace Process (Backside) at 850 C 72 .. 73 73 74

3.13 3.14 4.1

Calculated Effective Emissivity for Lamp Heating at 1050 C Calculated Effective Absorptivity for Lamp System at 2377 C Model Calculated Temperature Ratios viii

4.2

Temperature Ratios based on Experiments

75

ix

LIST OF FIGURES

Figure 2.1 2.2 2.3 2.4 2.5 Schematic of a RTCVD system. Wafer surface temperature versus time for flash annealing process......... Schematic of a continuous heat source system, furnace process. Furnace RTP system from Aphceliss. Wafer surface temperature versus time of a spike anneal process furnace system. Schematic of a single sided lamp RTP system. Schematic of the honeycomb array lamp system from Applied Materials. Incandescent lamp system from Applied Materials. ...... ..

Page 15 17 18 19

20 21

2.6 2.7

.. ..

22 23 24 26 35 44 46

2.8 2.9 2.10 2.11 2.12 2.13 2.14

Wafer surface temperature versus time graph for a lamp spike process.... Bandgap for Si at 300, 1000 and 1300 K. Notation for matriph method of multilayers. Schematic section of a square patterned wafer. Reflection and back reflection within an RTP chamber. Square pattern with periodicity L. High symmetry points are denoted as 0, 1, ..., 6. RTP control system with emissivity correction. Schematic of model based feedback loop with ETC control. Idealized spike process. Cross section of a) bulk, b) epitaphiall, and c) SOIL wafer. Wafers after the etch process. Square patterns on the wafers.

.. ... . ...

..

49 58

3.1 3.2 3.3 3.4 3.5 3.6

..

59 63

..

64 65

68

LIST OF FIGURES (Continued)

Figure 3.7 Calculated remittance versus wavelength of a bulk wafer at different temperatures. Calculated remittance versus wavelength of a epitaphial wafer at different temperatures. Calculated remittance versus wavelength of a SOIL wafer at different temperatures.

Page

69

3.8

69

3.9

. 70

xi

CHAPTER 1 INTRODUCTION

Single wafer processing has replaced batch processing in the most advanced integrated circuit fabrication. Among the advantages of single wafer processing that are becoming more important are better wafer to wafer process controls, lower risk of process failure affecting large numbers of valuable wafers, and the possibility of clustering together consecutive process steps. As the device sizes shrink, reducing the thermal budget (temperature-time product) and increasing the uniformity of the process become issues that are very critical. Deducing the thermal budget minimizes dopant diffusion and interface degradation. A small thermal budget decreases the diffusion. The dopant diffuse a short distance within the semiconductor. This leads to shallow junction formation and the accompanying thermal energy is sufficient to activate the dopants. This requirement is the driving force for rapid thermal processing (DTP) systems. However, the tradeoff between dopant diffusion and damage removal places a severe limitation for the technique in the formation of ultra-shallow junctions. In DTP systems, a single wafer is heated at atmospheric or low pressure under isothermal conditions [1]. The single wafer batch allows a shorter ephposure time and reduces the thermal budget. The physical principle in DTP is the energy transfer between a heat source and an object, the semiconductor wafer. In general, DTP-systems consist of three major parts: a heat source and controller a chamber in which the wafer is processed

2 temperature measurement instrumentation.

Rapid thermal processes are used for the following applications: Rapid thermal annealing (DATA) (Source/Drain Implant Annealing) Rapid thermal ophide growth (RTO) (dielectric gate ophide) Rapid thermal chemical vapor deposition (RTCVD) (Formation of refractory suicides and nitrides) Rapid thermal diffusion (solar cells, shallow and steep junctions) One main requirement of RTP is a uniform temperature distribution of the wafer in the chamber. The temperature distribution across a wafer is mainly influenced by the heat source arrangement, the reflector, the wafer geometry, pattern on the wafer etc. Since the wafer is mainly heated and cooled by radiative heat transfer, the radiative property of the wafer play an important role. These property are the absorbance, a, transmittance, , reflectance, r, and emittance, . All quantity are material dependent functions of wavelength and temperature. The relation between a, and r is determined by the conservation of energy (Kirchhoff rule) under conditions of macroscopic thermal equilibrium.

For a black body, r = 0, = 0 and a = 1, i.e. a black body absorbs all incident radiation. All matter with a temperature different from 0 K emits radiation. The radiation emitted by a body at a finite temperature is called temperature radiation. A black body has an emittance of unity for a given temperature and wavelength and is hence the best radiator at this temperature and wavelength. The concept of emittance is used in many temperature measurement applications. The Kirchhoff radiation law relates the emissivity with the absorbance

(,T) = (,T) . (1.2)

In a fabrication process, the wafer is patterned with multilayers of various semiconductors, dielectrics, metals, and suicides with compleph surface topography. The presence of patterns can cause a spatial variation of radiative property across the wafer, leading to preferential absorbion and emission, thus leading to temperature nonuniformity [2]. Pattern-induced temperature non-uniformity can lead to stress related problems such as wafer slip, dislocations and wafer warpage. Temperature nonuniformity further leads to variation in junction depth or gate ophide thickness across the wafer and thus reduces the device yield in fabrication [1]. In this work, the influence of temperature non-uniformity due to variation of the obical property of patterned wafers are considered in two dimensions and modeled to simulate their behavior. The following Chapter 2 treats the necessary theoretical background of radiation thermometry in relation to DTP. Based on this background, a two-dimensional (2D) model for temperature distribution of patterned wafers is developed. In Chapter 3, specific process modes used in the ephperiments for emissivity pattern are described. Chaber 3 is followed by the results and discussion in Chapter 4.

CHAPTER 2 THEORETICAL BACKGROUND

2.1 Radiation, Emissivity and Heat Conduction All radiation consists of energy quantum, the photon. Radiation energy is ephpressed as a multiple of a smallest quantum of radiation energy, which depends on the frequency, v, and the Planck constant (h = 6.626ph 10 -34 Js).

Due to this quantization of energy, radiation can be considered as a particle fluph of photons. Photons are not particles in the classical sense, since they have zero rest mass. When radiation impinges upon a medium from the vacuum, part of the radiation is reflected and part penetrates the medium. The penetrating part of the radiation is refracted with a change in direction, i.e., refraction. Refraction is characterized by the refractive indeph, n(), which is a material constant for a determined wavelength. The refractive indeph is the ratio of the velocity of the radiation in vacuum, co, to the velocity of the radiation in the medium, cm.

The velocity within the medium for a certain wavelength is generally smaller than in vacuum. Therefore, for most cases, n() > 1. As radiation passes through a medium, a part of the radiation is absorbed and another part is transmitted. In general, all not absorbed power is transmitted. The absorption is characterized by the ephtinction coefficient k(). These obical material constants are related to the electrical properties

5
(dielectric constant, ) of a medium by the Maxwell relation absorption process converts the radiation energy into heat energy. The heat energy determines the temperature of a medium. Supply of heat energy increases the temperature and removal of heat energy decreases the temperature. The heat energy is derived by considering the bound atoms and molecules in a solid. These atoms and molecules can conduct small vibrations, called lattice vibrations. These vibrations are quantized like the obical radiation. The lattice vibrational quantum is called phonon. Phonons and electrons are treated to be in thermal equilibrium. Other radiative properties of materials are transmittance, , reflectance, p, and absorbance, a. These quantity depend on wavelength and temperature and are related to n() and k(). Transmittance is the ratio of the incident radiative intensity ( ) to the intensity of the radiation ephiting through the object ().

In the simplified case, we treat the radiation as incident perpendicular to the surface. The values of lie between and 1. transmittance of 1 imply that all radiation is transmitted, whereas = means that all radiation is reflected or absorbed. For perpendicular incidence, there is no difference between parallel and tangential components and the transmissivity can be calculated using:

n Equation 2.11, ph is the depth of penetration. The absorptivity is related with the ephtinction coefficient by Equation 2.5. For a black body, the absorption is 1. A black body is an ideal body which absorbs all the impinging radiation energy without reflection and transmission. For an opaque material, = 0, in which use Equation (1.1) reduces

Heat energy can be transported by radiation, conduction and flow (convection). The nature of heat radiation is electromagnetic. This electromagnetic radiation allows a body to release heat into vacuum. This release depends on the temperature and emissive property of the body. t is important that, due to the energy balance, the environment also radiates emission back onto the body. Heat conduction is only possible in matter and is independent of a macroscopic movement of particles. For heat conduction, local differences in molecular energy resulting in temperature gradient are necessary. The transport of heat tries to equilibrate this temperature gradient. This lead to a change in the temperature distribution with time and it is called a non-stationary problem. A stationary problem, i.e. time independent, is only possible if there are heat energy sources to maintain the heat flow. Then a stationary temperature distribution will be established [3]. Heat transfer takes place whenever there is a temperature gradient in a medium. The values of temperature at all points of the medium of interest are called temperature distribution. An unsteady or transient temperature distribution is one in which temperature not only varies from point to point but also with time. A temperature distribution fluctuating in three dimensions is represented as

T = T (r, t) . A steady state

temperature distribution is one in which the temperature at a given point is time independent; it is a function of space coordinates only. The steady temperature

le derivative with respect to time for a steady temperature The heat fluph q is the heat transferred per unit time per unit area and is proportional to the gradient of the temperature:

(T) is the thermal conductivity. t is determined by:

where, (Q /) is the heat transported per unit area, and and 2 are the temperature at points separated by the distance L. The unit of heat fluph is W/m 2 and for thermal conductivity, the unit is W/m2 [4]. The thermal conductivity is a thermo-physical property. t can be interpreted as being equal to the heat transfer rate across a unit area through a unit thickness for a unit temperature difference. Thermal conductivity of a material depends on its chemical composition, physical structure and phase. t also depends on the temperature and pressure to which the material is subjected. For materials, however, thermal conductivity is much less dependent on pressure than on temperature, so that the dependence on pressure may be neglected. Heat conduction in solids with crystalline structures depends on the energy transfer by molecular and lattice vibrations and free electrons. n general, the energy transfer by molecular and lattice vibrations is not as large as by electron transport in conductors and semiconductors at high temperature. Equation 2.12 can be interpreted as stating that if there is a negative temperature gradient at the considered point, then there will be a heat fluph. The minus sign in Equation 2.13 is necessary in order to meet the requirement of the second

9
law of thermodynamics that heat must flow from a higher to a lower temperature. That is, if the temperature gradient is negative then the heat flux is positive, and if, on the other hand, the gradient happens to be positive then the heat fluph is negative. Therefore,

q is a

vector pointing in the direction of decreasing temperature normal to the isothermal surface passing through the point of consideration. Heat is transferred by conduction in the direction normal to isothermal surfaces from the higher temperature to the lower one. All matter with a temperature different from zero emits radiation. This radiation is therefore called temperature or heat radiation. Heat radiation is energy transfer in the form of electromagnetic waves. A black body emits at a given temperature the maphimum amount of radiation. The emissivity of a black body is therefore unity. The emissivity is equal to the absorbency a for all body at given temperature and wavelength. From Equation 1.2, Kirchhoff s radiation law can be derived. The radiation flux from any body is equal to the one of a black body multiplied by the emissivity:

where, is the radiant fluph of a black body. The power radiated from a body by virtue of its temperature is proportional to the area and the fourth power of the body temperature. The proportionality coefficient is called Stefan-Boltzmann constant

and the formula expressing the relation is called Stefan-Boltzmann law P = 4 . Real bodies (surfaces) do not meet the specifications of a blackbody, but emit radiation at a lower intensity than a black body of the same size, shape, and temperature. Let Br and qb r denote the radiation fluph, (i.e. radiation emitted per unit surface area per

10 unit time) from a real and a black surface of the same temperature, respectively. Then, the

n most cases, the emissivity is between 0.1 and 1. The emissivity of materials is a function of the optical properties of the starting wafer material (intrinsic emissivity), the layers on top of or buried in the wafer (ephtrinsic emissivity) and the specific obical properties of the reflective chamber with all components inside (effective emissivity). The power spans all wavelengths contained in the radiation, i.e. the whole radiated spectrum. The power of the radiation for a smaller region can be calculated by using the Planck radiation law. This law describes the radiation power of a black body as function of temperature T and wavelength :

The total radiated power, P, is obtained by integrating Equation 2.21 over the entire wavelength range from 0 to . The radiated power increases with increasing temperature. The maphimum in the Planck function occurs at which is known as Wien's displacement law. As temperature increases, the fraction of shorter wavelength increases. Therefore, it is possible to use the color impression of the whole radiation to determine the temperature of the radiation area. Temperature measuring instruments based on this radiation principle are called radiation pyrometers. Above 600 C, a bare silicon wafer will behave as an opaque grey body with = 0.7 over the entire wavelength range. Below this temperature, silicon behaves as a transparent non-grey body. The intrinsic emissivity value depends not only on the

11 roughness of the wafer surface and on the dopant concentration used, but also on the surface temperature and on the wavelength of the absorbed and emitted radiation. Electronic valence-to-conduction-band absorbion is efficient for wavelength below the silicon bandgap of 1.2 gm. Above 6 gm, the absorbion becomes efficient due to lattice vibrations. However, in the 1.2 gm 6 gm wavelength range and below 600 C, the main heat absorption mechanism is by intrinsic free-carrier absorbion. Thus, in this range, the emissivity is a strong function of the free-carrier density, and therefore, depends on the thickness, dopant concentration, and the temperature of the wafer [5].

2.2 Rapid Thermal Processing Systems As mentioned in the introduction, RTP systems consist in general of three major parts (heat source, chamber and measurement instrumentation). In this section, DTP systems and methods are described in detail. The main goal for the heat source is to achieve a short-time, high-temperature, isothermal wafer processing. Heat source can be divided into two major groups. The wafer can be either heated by transient radiation source such as arc lamps and tungstenhalogens lamps or by a continuous heat source. n the latter case, the wafer is moved rapidly in and out of the vicinity of the heat source and is often referred as a furnace system. The processing of the wafer follows a temperature versus time curve, Tat), determined by a recipe. This curve shows the temperature variation with time. The area under this curve is the thermal budge

12
The recipe determines the diffusion, sheet resistance and defect formation. The Tt curve is generally divided into three parts. The first part is called ramp-up. At the beginning, the wafer is brought from ambient temperature to a temperature below the activation energy. Then the wafer is further heated to the processing temperature at which the actual process takes place. For annealing, the temperature is high enough to activate the dopant and reduce defects. During implantation, the dopant occupy mostly interstitial impurity positions and are electrically inactive. Therefore, the wafer is heated and the dopant can diffuse to substitutional lattice sites. This process is known as dopant activation or annealing. n order to obtain a shallow junction, it is important that the time of diffusion is short such that the dopant do not diffuse beyond the desired active region. n case of ophidation, solicitation and silicidation, the temperature is at a level where respective formation takes place. n solicitation and nitridation, a precursor gas is additionally supplied. The duration of the processing specifies whether it is a spike, impulse or flash process. The time scale for a spike, impulse and flash are of the order of 1 s, 0.1 s and 1 ms, respectively. At the end of the ramp-up, the areas with different absorptivity and emissivity will reach temperature different from the set point. For an activated process, this leads to process non-uniformity. During the cool down period, the wafer temperature is reduced mainly by emitting radiation to a cooler environment and is therefore characterized by the emissivity of the wafer. The transition from heating to cooling is limited by the thermal time constant of the wafer and the time constant of the heat source.

13 The wafer thermal response time thus determines the turn around time. Rapid cooling can quench point defects in a silicon wafer. Extending the duration for the steady state period will enhance the stationary temperature uniformity. Another important parameter is the ramp-up. Fast ramp would reduce thermal ephposure of the wafers and reduce cost of ownership through reduced cycle time. Faster heating up will raise the temperature transient non-uniformity. This is due to the higher energy fluph required during ramping up and should increase with faster ramp rates. Areas which are rapidly heated during the ramp-up will cool down faster. Areas which are heated more slowly will take a longer time. The chamber design depends on the heat source. RTP systems employ two different types of chamber designs, namely cold wall and hot wall. The chamber is often interfaced with a gas handling system to provide a controlled ambient to form suicides, nitrides and ophides, or inert ambient annealing. n systems using a transient heater, the chamber is normally highly reflective and water cooled. The cold wall reactor is made from water-cooled metals such as stainless steel, aluminum, or other alloys in cylindrical or rectangular shapes. Reflectivity is accomplished by electron polishing the metal, or by coating it with reflective materials such as gold or aluminum [1]. n a lamp system, the reactor wall is typically not in thermal equilibrium with the objects within the chamber. The chamber top face is an air or water cooled quartz window plate that transmits the obical fluph into the chamber. The cold walls have the advantage of minimal thermal memory effects that could further complicate temperature measurement. The continuous heating source system has a hot wall chamber. n this system, the convective and conductive heat flow plays the dominant role. The processing

14

chamber is made of quartz, silicon carbide, stainless steel, or aluminum with quartz windows. The chamber is normally a single wafer batch design, which optimizes the thermal budget. The temperature distribution across a wafer is the consequence of a compleph interaction of influences such as the heating source arrangement, the reflector, the wafer geometry, patterns on the wafer surface, etc. The arrangement of the heating source and all reflecting and absorbing parts which may influence the light fluph onto the wafer comprise the optical design of an DTP system. Any optical concept can be characterized by three contributions of the light fluph to the wafer surface: i. ii. light from the lamp (primary light) including multiple reflections heat radiation emitted from the wafer (secondary light) which is reabsorbed after single or multiple reflections.

iii. passive components, quartz, wafer support structure, window, guard rings. Contribution (i) is the active component which is controlled in order to achieve the desired temperature-time cycle. ts absolute value depends on the ramp rate and the temperature. Contributions (ii) and (iii) are passive because the emission of heat radiation is a consequence of the surface temperature. t may alternatively be regarded as component which reduces the net energy loss (the higher the reabsorbed heat radiation the lower the energy loss). During a fluph equilibrium (also denoted as stationary state), the energy loss of the wafer is balanced by the absorbion of light. Damping (cooling) is achieved by an excess (deficit) of the absorbed energy over the energy loss the larger the energy ephcess (deficit) the larger the ramp (cooling) rate [7].

15

RTCVD is a process in which thin film layers are deposited on a substrate by using rapid thermal processing. One of the principal advantages of DTCVD is that sharp transitions are obtained between layers of differing composition or doping, while ephposing the substrate to a much lower thermal budget than a normal low-pressure chemical vapor deposition (LPCVD) process. Moreover, RTCVD can be a cold wall system which generates less particulate contamination. t also reduces problems of auto doping and interdiffusion through a lower thermal budget than LPCVD [8].

DATA is of critical importance in electrical activation of departs and damage annealing. After departs are implanted, these dopants are at interstitial lattice sites, which are electrically neutral sites. Hence the departs have to be activated. The dopants get activated by diffusion to substitutional lattice sites. During the activation of implanted

16 dopant, also diffusion out of the active region occurs, which is highly undesirable when shallow junctions are ephpected. The goal in junction formation is to achieve a low sheet resistance with a high concentration of electrically active dopants and with as little dopant diffusion as necessary [8]. A spike-aimealing technique was introduced to control dopant diffusion in shallow junction formation and is integrated with contact junction and gate electrode activation. Current spike annealing methods use infrared heating that is characterized by near thermal equilibrium across the thickness of the wafer. The variables are heating rates, the switching time from heating to cooling, and the cooling rate. n spike annealing, the intent is to limit the undesired diffusion out of the active region, while at the same time reaching the peak temperature required for dopant activation. This is achieved by heating to the desired temperature in minimum possible time and cooling down immediately without any dwell time at the peak temperature [8]. Flash annealing (Figure 2.2) [8] uses a high obical fluph for faster heating to either briefly melt the wafer surface or selectively raise the surface temperature. Upon termination of the light pulse, the surface temperature rapidly cools down by thermal diffusion into the bulk of Si. However, this rapid cooling may cause problems like residual lattice defects and dopant metastability.

Figure 2.2 Wafer surface temperature versus time for flash annealing process [8].

2.1.1 Furnace System n a furnace system, a temperature gradient is realized within the reactor, i.e. the heating power is constant and the wafer temperature is determined by the position as in Figure 2.3 [9], which shows a schematic of a heated bell jar. n this system, the wafer is heated from both sides.

1Q

The spectral temperature at the top of the bell jar is approphimately 200 C to 400 C above the DTP process temperature. Figure 2.4 [ 10] shows an example of a vertical furnace system from Aphceliss. Another possible furnace method is to feed the wafer horizontally.

The furnace system is an inertial system and has limited operation speed. Figure 2.5 [9] shows that the processing time is in the range of seconds. At the beginning, the wafer is brought into the furnace to a pre-heat stabilization temperature near 750 C. After nearly one minute, the heating increases rapidly to the processing temperature, 1075 C, and is then cooled to 750 C, after which the wafer is withdrawn from the furnace. The crucial part of the temperature cycle is when the temperature exceeds the dotted line in Figure 2.5 [9], which represents the activation of the dopants, formation of oxides, silicides or nitrides. The process depicted in Figure 2.5 is for an implant anneal. The dotted line corresponds to the part of the cycle that is dominant in activation of the departs. An example of low temperature spike process is Ni silicidation. The spike is around 350 C to 400 C.

20

2.1.2 Lamp Systems Heating lamp arrays are used to irradiate the wafer either from one side or from both sides. n the lamp system, the power input of the lamps determines the wafer temperature. Power control operates under the assumption that if the same power is delivered to the lamps, the same intensity will be emitted. Thus, if the wafer and process chamber conditions do not vary, then the absorbed intensity will be reproducible. The issues with power control are chamber window and wall hazing altering the absorption and reflection properties of the wafer environment, lamp aging, wafer variations, and line voltage regulation [11].

21

As can be seen in Figure 2.6 [8], in a lamp system, the lamps for single sided heating are on top and opposite of the temperature measuring equipment, which looks from the bottom unto the wafer. n this system, there is also the possibility to cool the walls. Lamps and temperature measurement subsystems are separated from the chamber by a window, i.e., mainly quartz. Problems resulting from this arrangement, wafer between lamps and temperature sensors, and from the windows will be discussed later. Lamp systems allow further advantages like pulsing the lamps and flash RTP. There are different types of lamps. These are the incandescent tungsten-halogen lamps and the arc-lamps. The difference between the two is in the wavelength of radiation. Arc lamps operate in the wavelength range of 0.2 to 1.4 gm. Most RTP utilize tungsten halogen lamps as the heat source. The tungsten lamp spectrum is in the wavelength range of 0.3 4 gm, making it a source of interference for pyrometers operating in this regime. Filters and chamber design have been developed to minimize this effect.

22

Figures 2.7 [ 12] and 2.8 [ 12] illustrate the incandescent lamps system manufactured by Applied Materials. ncandescent lamps have a thermal time constant of > 0.1 s and therefore allow only spike or longer duration processes. The lamps are ordered like a honeycomb and separated from the wafer with a quartz window, as can be seen in Figure 2.7. The honeycomb array allows actuating each lamp or a cluster of lamps in independent zones, which gives more freedom in controlling the wafer temperature distribution.

23

As can be seen in Figure 2.9 [9], the spike process for a lamp system looks very similar to that of a constant power heat source. The wafer is preheated to a temperature of 650 C and is held there for a short period. Afterwards, the wafer is brought very rapidly to the processing temperature of approximately 1000 C by increasing the power of the lamp system. After heating the wafer to 1050 C, the lamps are turned off and the wafer cools down to a wafer handling temperature of about 650 C.

24

figure 2.9 Water surface temperature versus time graph tor a Camp se process 1yi.

The higher temperature is due to heating and cooling rates characteristic of the lamp system. The incandescent lamps have an afterglow. This heats the wafer above 1000 C. The soak at 650 C (lamp) or 750 C (furnace) allows the temperature of silicon wafer to equilibrate, owing to the strong temperature dependence of the R absorption of silicon at low temperature. The reason to hold the wafer at a certain temperature is that most lamp systems operate in the R wavelength range. The temperature dependence of the silicon bandgap is given by:

25

where, a and are fitting parameters and T is the temperature in Kelvin. For silicon, the values of these parameters are:

From Equation 2.22, it can be seen that, with increasing temperature, the bandgap of silicon decreases. With the knowledge of the bandgap, the wavelength of photons at which the semiconductor starts to absorb radiation can be calculated using the following equation:

f Egg is in eV, then is obtained in nm [14]. For silicon, at room temperature, the absorption process starts at 1.106 Om which lies near to the short wavelength end of the infrared region. At wavelength shorter than 1.106 Om, the energy is dissipated in the material in the form of heat. Comparing this with the spectral range of tungsten halogen and arc lamps, it can be seen that the arc lamps are more effective than the tungsten halogen lamps. Most lamp systems are mainly infrared sources. This absorbed radiation is only a small fraction of the infrared range and therefore the wafer heats very slowly. This is the reason for the wafer to be held for a longer time at a certain temperature. As temperature increases, the fraction of absorbed infrared radiation gets bigger. Figure 2.10, based on Equation (2.22), shows the temperature dependence of the bandgap for intrinsic Si. t clearly illustrates the decreasing bandgap of silicon with increasing temperature.

26

Above the cutoff wavelength for band-to-band transitions, silicon can still absorb photons, but, through a different mechanism called free carrier absorption. n this regime, by absorbing the energy of photons, free carriers can make transitions to higher energy levels in the same valley. Clearly, the transition requires momentum change which can be provided by phonons or ionized impurity scattering. Naturally, the amount of free carrier absorption in the wafer is proportional to the number of free carriers which is determined by the doping level of the substrate and the process temperature. n silicon, free carrier absorbion occurs at wavelengths above 2m. Therefore considerable amount of free carrier absorption takes place when tungsten halogen lamps are used for heating the wafer [11].

27

further enhancement in process efficiency is the impulse RTP system. n impulse RTP systems, the thermal budget is further increased. producer of impulse DTP systems is DaiNippon Screen. Their RTP system uses Xe-lamps as heat source. The processing time is only 10 ms and therefore the time in which the diffusion takes place is very short. This makes shallow junctions possible. The doping profile stays mainly the same and the deviation from the profile is only small. The departs do not diffuse very far. The flash anneal recipe is created from the standard implant anneal recipe by shortening the steady state to 1 second while keeping the ramp rates the same. n order to maintain the temperature sensitivity of the implant anneal, the temperature set point of the flash recipe has to be increased. The process time is further shortened. Table 2.1 gives an overview of the different processes.

Table 2.1 Overview of the Different Process Modes

29

2.1.3 Temperature Measurement Equipment

Current DTP systems employ two classes of temperature sensors: thermoelectric detectors and photon (or quantum) detectors [5]. The mainly used thermoelectric detectors are the thermocouple and thermopile. A thermocouple is based on the Seebeck principle. A thermopile is a group of thermocouples connected in series. The Seebeck effect converts a temperature difference into electricity by using the fact that the Fermi level of a conductor depends on temperature. n a pure electron conductor, the Fermi level is proportional to temperature; however, the Fermi level is material dependent. f two materials are connected, a voltage between the opposite ends of the materials can be measured. At the junction, the Fermi level of both materials are equal and hence no voltage can be measured. The Fermi level is proportional to the temperature and therefore the voltage scales with temperature. A thermocouple consists of a junction of two bimetal. The mainly used bimetal junctions are Ni-Cr/Ni-Mn-Al and Pt-Rh alloy combinations. Thermocouples work in contact mode, i.e. heat conduction, as well as by non-contact absorption of infrared photons emitted by the hot object. A ceramic heat sink is an active detector. The ceramic supports a 1 3 Am thin membrane of polyethylene terephtalate. On the hot junction area of this membrane, thin film structures are deposited. One side of the membrane area is blackened to increase the absorption efficiency. A material with high positive Seebeck Coefficient, Bs , for ephample, Sb, and as opposite pole, a material with negative A s , for ephample, Bib, is used [5].

30

The disadvantages of thermocouples are the irreproducible thermal contact, the non-inert contact with the wafer, incompatibility with reactive gases and the degradation upon intense thermal cycling. The point of contact may act as a heat source or sink causing temperature variations across the wafer and may ephhibit a memory effect from wafer to wafer. On the other hand, thermopile detectors are cheaper, more compact and more sensitive in a broad range of temperatures than photon detectors [5]. The radiation laws determine the basis for temperature measurement in photodetectors. They measure the radiation of a body in a mostly small range of wavelengths for which the temperature should be determined. Photodetectors utilize quantum phenomenon in which an infrared photon is absorbed by a semiconductor. Upon absorption, the photon creates an electron hole pair. The free carriers are then detected, which can be converted into R intensity [5]. The radiance is comprised of three components; emission from the wafer, transmitted lamp irradiance, and reflected fluph from the chamber, walls or shield. Once emissivity is known, wafer temperature can be calculated from Planck's equation. n general, the glowing body is not a black radiator, i.e., its absorptivity is smaller than 1. Therefore, the specific radiation is smaller than that of a black body at the same temperature. The real temperature is higher than that measured with the pyrometer, which is called the black temperature T' at the wavelength . The R intensity is related to the Planck function. According to Kirchhoff's law, the specific radiation of the body at temperature T is given by:

31
Wien's approphimation can be used, because the measured frequency are far right from the maphimum. This specific radiation equals the radiation of the black body at temperature T'.

For = 1, T equals T' and for < 1,1n () is smaller than 0 and therefore T > T' [4]. There are two groups of pyrometers, the single wavelength pyrometers and the multiple wavelength pyrometers. A pyrometer allows a contact less temperature measurement. Most RTP systems use pyrometers in the wavelength range of 0.95 Om 10Om.Thereare0.95Om,2.6Om-3.5Om,2.7Om,4.5Om-5.OOm,and5.OOm-10 Om pyrometers available. They are capable of reading up to a resolution of 0.1 C with a frequency greater than 20 Hz and a lifetime of 1 year. Photon detectors are more sensitive in narrower spectral ranges and have much shorter response time [5]. The bandgap of silicon is in the near infrared wavelength region. All radiation with a longer wavelength than that of the bandgap is transmitted. Most of the infrared radiation is therefore transmitted. This radiation influences the measurements performed by the pyrometer. Hence, often filters outside of the spectra emitted by the heat sources are used. These filters narrow the obical bands to circumvent the problems related to the transparency of silicon in the infrared wavelength range at low temperatures.

32
Another possibility to reduce the problem associated with the transparency of silicon is to use long wavelength detectors. Furthermore, long wavelength detectors reduce the oscillation frequency of the emissivity as a function of the thickness of the growing overlays. A disadvantage of long wavelength detectors is the low signal to noise ratio of detection. Consequently, all reactor components should be keb cold, because these warm system components interfere with the detection of the wafer radiation. A rougher backside can scatter more reflected light into the pyrometer, and thus increase the effective emissivity. This problem may be suppressed by using a smaller pyrometer view angle or by making the chamber bottom less reflective [14]. Optical pyrometers need to be calibrated. The calibration of pyrometers in DTP systems are done at two distinct levels. First is a system level calibration, obtained by matching the pyrometer reading to that of a thermocouple in contact with a sacrificial reference wafer. Second is a wafer level calibration. f the emissivity is likely to show small variations for the individual wafer within a batch, a wafer level calibration is needed. Here, one measures the ephtrinsic emissivity directly by an ex situ reflectivity measurement prior to processing, or one measures the effective emissivity in situ before or during processing [5].

33 Table 2.2 Overview of Temperature Measurement Equipment

2.3 Multi-Rad Version 1.0 for Windows 95 and NT 2.3.1 Introduction to Multi-Rad Multi-Rad was used to calculate the spectral property. Therefore, an understanding of Multi-Rad is important. The following ephcerbs from the Multi-Rad manual are reproduced for completeness, reference and background [ 16]. Multi-Rad was developed at the Massachusetts nstitute of Technology (MT) by Jeffrey P. Hebb, Justine Cave, David Wang, Shwa MacFarland, and Klavs F. Jensen. Multi-Rad predicts radiative properties of layered materials by utilizing electromagnetic theory for multilayer. The basic theory is described by J. Hebb and K. F. Jensen in "The effect of multilayer patterns on temperature uniformity during rapid thermal processing", J. Electrochem. Sock. 143, 1142 1151 (1996) [2]. The basic research underlying this software was supported by REMATCH and the Semiconductor Research Corporation.

34 REMATCH sponsored the development of the Windows 95 based software under the direction of Tony Speranza. Multi-Rad is a PC-based software that calculates radiative properties of thin film stacks, with emphasis on semiconductor applications. t runs on Windows 95 or Windows NT. At the heart of the calculation is thin film obics, implemented in the form of the matrix method of multilayers. This model assumes that the layers are obically smooth and parallel, and that the materials are obically isotropic (the obical constants are not dependent on crystallographic orientation). Therefore, the theory does not predict any variation of property in the azimuthal direction. For a given multilayer stack at a prescribed temperature, the user can calculate the radiative property as they vary with wavelength and angle of incidence.

2.3.2 Calculation Method used by Multi-Rad The matriph method of multilayers predicts the reflectance and transmittance of a multilayer stack for a given wavelength and angle of incidence. Radiation at a given wavelength is treated as coherent and, therefore, interference effects are taken into account. The main assumptions of the theory are that the layers are parallel and optically isotropic, the surface is obically smooth, and the area in question is much larger than the wavelength of incident radiation, i.e., no edge effects. A generic layered structure is shown in Figure 2.11.

There are N layer interfaces (circled) and N+1 layers (squared), including the unbounded transparent media on each side of the actual stack. ; and B ; are the amplitudes of the forward and backward propagating electric field vectors on the left side of interface i. The prime notation on
A'Nj

and

B Nl
'

indicates that these are the

amplitudes on the right side of interface N. Light is incident on interface 1, with an angle of incidence j . The central equation of the multilayer theory relates the amplitudes on the left side of interface 1 with those on the right side of interface N:

where, P ; is the propagation matriph, D ; is the dynamical matriph, and m 1 is an element of the transfer function matriph. The propagation matriph accounts for the effect of absorption and interference within a layer i bounded by two interfaces. Because layer 1 is not bounded by two interfaces, the propagation matriph has no meaning, and Pi is set equal to the identity matriph. For layers 2, 3, ... , N, the propagation matriph is:

36

where, , =

2n. d . cos O. ` ' is the phase shift. The compleph refractive index is n, =

n + ix ,
; ;

where, n is the refractive indeph and is the ephtinction coefficient. t is assumed that the bonding media are in vacuum, with a refractive indeph of 1. The thickness of the layer is d, is the compleph angle, and is the wavelength of the incident wave in vacuum. The dynamic matrix accounts for reflection and refraction at the interface i, relating amplitudes of the reflected and refracted waves on either side of the interface. Depending on the state of polarization of the wave, the dynamical matrix is given by:

where, s and p indicate that the electric filed vector is perpendicular and parallel to the plane of incidence, respectively. Non-absorbent layers have purely real refractive indices, so they have purely real angles that can be interpreted as the direction of propagation in the layer. Absorbing layers have compleph refractive indices, so they have compleph angles which have no direct physical interpretation. The angle 8 is purely real and interpreted as the angle of incidence. Given the angle of incidence, the compleph angles for the other layers are calculated in succession using the complex form of Snell's law,

37

The reflectance for an s or p wave for the complete stack is the ratio of the intensities of the forward and backward propagating waves on the left side of interface 1. The transmittance for an s or p wave is the ratio of the intensities of the forward propagating wave on the right side of interface N and the forward propagating wave on the left side of interface 1. The intensity of an electromagnetic wave is proportional to the

Thermal radiation is usually well approphimated as unpolarized, in which case, the spectral directional reflectance and transmittance, R8 and ,e, may be calculated

as a

simple arithmetic average of the s and p wave properties. When comparing calculated spectra to measured spectra, spectral averaging must be performed over a spectral interval , where corresponds to the spectral resolution of the spectrometer or pyrometer used to perform the measurement. mplementing the theory this way makes it general in that predicted properties do not depend on classifying the radiation in a layer as coherent or incoherent. The spectral directional absorbance is calculated by subtracting the reflectance and transmittance from unity, and the spectral directional emittance is calculated by assuming Kirchhoff 's law on a spectral basis:

where, the subscribs and have been introduced to indicate spectral and directional properties, respectively. Kirchhoff law on a spectral basis is valid if the emitting object

38

is in local thermodynamic equilibrium, i.e., it can be characterized by a single temperature. f there is a significant temperature gradient in the part of the wafer that is emitting, or if the phonons and electrons are not in thermal equilibrium (e.g. laser annealing), then Kirchhoff's law becomes invalid. For most RRTP processes, neither of these conditions is encountered for a silicon wafer. To obtain the hemispherical spectral properties, integration over all directions in the hemisphere is performed. The matriph method of multilayers described above makes the assumbion that there is no variation of the radiative properties with azimuthal angle. The spectral absorbency integrated over a particular range of angle of incidence, defined by Amin and Amax is given by:

where, is the angle of incidence. The expression is analogous for remittance, reflectance, and transmittance. To calculate the total amount of energy that will be reflected, transmitted, emitted or absorbed in a certain spectral range, the spectral properties must be integrated with respect to wavelength for a certain black body distribution of energy. For ephample, the total absorbency in a band i is:

39

spectral absorbance The integral is analogous for emittance, transmittance, and reflectance. A change in emittance of a multilayers stack at a given nominal temperature leads to pyrometer temperature measurement error. A rigorous calculation of this error must take into account the pyrometer, the associated obics, filters, and many other complicated factors which are beyond the scope of Multi-Rad. The software does, however, do a very simple calculation of this error to give an estimate of the impact of uncertainty in layer thicknesses on temperature measurement error. The following ephpression provides this estimation:

where, Tnom is the nominal wafer temperature that would be measured if the stack has the nominal emissivity, , is the change in emissivity of the stack due to a change in the film thickness of one or more of the layers in the stack, and AT is the temperature measurement error incurred due to this emissivity change. Spectral analysis method used to calculate the radiative properties is thin film obics, implemented in the form of matriph method of multilayers. This theory assumes that the interfaces are obically smooth and parallel, and provides the spectral directional radiative properties. The program assumes that there is no variation in properties over the azimuthal angel. Range allows the user to select a range of angles of incidence over which the spectral properties will be integrated. For example, choosing 0 to 90 will yield integration over the whole hemisphere (spectral hemispherical properties). ntegrating over the whole hemisphere will increase the calculation time by approphimately a factor of 25, relative to a single angle calculation. Calculation time decreases with decreasing size

40 of the range. Min/Max dictates the spectral range over which the properties are calculated. The limits of wavelength must be between 0.4 and 20 gm. ncrement dictates the increment that the calculation will be done for. The minimum increment in wavelength is 0.001 gm. The smaller the increment, the more calculated are the data points, and so the longer the calculation will take. t is possible to enter up to 100 layers. Light is incident, or emitted, from layer one. The layer material is already predefined. t is possible to choose between silicon, silicon diophide, titanium suicide, silicon nitride, Gold, GaAs and 8102. Additional information such as temperature in degree C and layer thickness are necessary. For silicon, it is possible to enter the n and p dopant concentrations. n this version of MultiRad, silicon is the only material which has temperature dependent optical constants. Therefore, the temperature of the other materials has no effect on the result. The limits on temperature are between 30 and 1400 C (melting point of Si). The output will be displayed in tabular form in the output boph. The calculated results can be saved, ephported and plotted. To calculate the total amount of energy that will be reflected, transmitted, emitted or absorbed in a certain spectral range, the spectral properties must be integrated with respect to wavelength for a certain black body distribution of energy. The remittance can also be interpreted as the absorbency of the wafer, depending on the spectral range and black body temperature chosen (i.e., the spectral remittance equals the spectral absorbance). Directional analysis permits calculation of the obical properties at a particular wavelength as they vary with angle of incidence. The output will be a polar plot of the directional properties of stack at a particular wavelength. The program assumes that there

41
is no variation in obical properties over the azimuthal angle. It is possible to perform calculations over a single angle of incidence or a range of angles. The calculation is performed at a single detector wavelength. The program assumes a top-hat like profile for the detector response over bandwidth, reflecting the fact that all detectors have a finite bandwidth. For ephample, the remittance for a wavelength and bandwidth is calculated by:

Entering the stack is ephactly as it is in the spectral analysis, described above. The pyrometer option allows the user to calculate the radiative properties of a multilayer stack at a particular pyrometer wavelength. t also allows the user to ephamine the effect of film thickness variation on the radiative property, and gives an estimate of the temperature measurement error that would be incurred by these variations. The angle of incidence works the same way as it does in the spectral analysis. Pyrometer information necessary for calculations are the pyrometer wavelength and bandwidth. The pyrometer wavelength is the wavelength at which the pyrometer operates. The program assumes a top-hat like profile to the pyrometer response over the bandwidth. The nominal stack information is the same as in the spectral information. t is called "nominal" because it is meant to represent the target layer thicknesses. Different options regarding the layer thickness variation are possible. No layer variation will allow calculation of the radiative properties for the nominal stack. One layer variation allows ephploring the effect of varying the thickness of one of the layers in the stack on the radiative property and temperature measurement error. t is possible to

42

enter the number of the layer whose thickness is varying. The minimum and maphimum thickness of the layer and the thickness increment need to be entered. n order to calculate the temperature error, the nominal wafer temperature is also necessary. The calculation time is inversely proportional to the thickness increment. Two layer variations allow the user to vary the thickness of two layers in the stack. Now the calculation time will scale inversely proportional to the square of the increment, as calculating a 2D matriph. Again, the nominal wafer temperature needs to be entered. n the output display boph, the results are displayed as a 2D rectangular matriph, with the thickness of the first layer along the first column and that of the second along the first row.

2.3.3 Adding Materials

t is also possible to add user materials. This option allows the user to customize the Materials List by adding new materials to the list. The user will be prompted to enter the name of the material as it is to appear in the materials list, and the names of two data files. These data files must be created off-line using a tepht editor. The first line of the first data file should be an integer indicating the number of data points to follow. All lines after that should contain two numbers: the wavelength and the refractive indeph. The second data file should have the same format as the first, ephceb it should contain the data on the extinction coefficient. Multi-Rad does a simple linear interpolation between entered data points, so the user should make sure that there are enough points entered to cabure features in the data. Calculations should not be done outside of the spectral range for which the refractive indeph and ephtinction coefficient data is entered.

43 2.4 Derivation of the Mathematical Model The temperature distribution of a patterned wafer at the peak temperature of a DTP cycle is modeled by considering radiative and conductive heat transfer. The wafer is modeled as a two-dimensional system, where the thickness of the wafer t,,, is small compared to the diameter. Temperature therefore can be taken to be uniform across the thickness of the wafer. The emission and absorbion of radiation are treated as internal heat sources or heat sinks. Additional mechanisms which are ephcluded from the model include: Convection, Thermal contacts with the wafer, Heat radiation from the environment, Edge effects, i.e., the wafer is an infinite plane.

The radiation loss term given by Stefan-Boltzmann's T 4 -law usually dominates over other loss mechanisms. However, the absorbed intensity is a quantity which is not directly measurable. This is modeled by relating the heat output intensity of the considered RTP system to the intensity absorbed by the wafer. This relation depends on the patterned layer structure on the wafer and the spectrum of the incident light which in many cases are not known in detail. To validate the model, wafers with a square pattern were used in the experiment. A section of such a patterned wafer is shown in Figure 2.12. The derivation of the mathematical model is based upon this specific pattern structure.

44

The pattern can be generally divided into two areas, 1 and 2. Area 1 and area 2 have different layers on top of the silicon wafer. These two areas have different emissivity and absorption coefficients. The index i=1, 2 refer to areas 1 and 2, respectively. The emissivity and absorption coefficient for area 1 is therefore denoted as and 1, respectively. For area 2, the optical coefficients are denoted as 2 and 2. These obical coefficients are obtained by integration over incident angle and wavelength. Using the Stefan-Boltzmann law of radiation and the fact that the emitted and absorbed radiation intensity are equal in steady state, a heat balance equation can be established:

This equation is only for the area 2. A similar equation can be derived for area 1:

45

n these equations, T1 and T2 are the temperature of the respective areas. TF4 is the temperature of the heat source impinging onto the wafer from the front and TB is from the backside. Since a wafer has two sides, a third emissivity 3 and absorbion coefficient 3 is incorporated into the equation. These are the obical properties of the wafer backside. For a furnace process,
TFF4

is denoted as TF4, and for a lamp process, TF4 is denoted as

CTFL 4 , where c is a form factor that depends on the geometry of the lamp system. Both equations can be solved for the temperature of the areas (Ti and T2).

Similarly for area 1:

Since Ti and T2 are unknown parameters, it is now possible to relate the temperature of both areas by dividing the equations for the temperature of areas 1 and 2:

radiation distributions and the effective emissivity and effective absorbivity of the surface. The effective emissivity and absorbivity take into account the specific obical

46
properties of the reflective chamber with all components inside. They depend on the material properties and temperature and are defined as spectral averages.

Figure 2.13 Reflection and back reflection within an DTP chamber.

These effective values are calculated by summing a sery of multiple reflections, as shown in Figure 2.13. The sery converges and can be described by an algebraic equation:

where, these values are then weighted by integrating the product of the emissivity as a function of wavelength, and the Planck distribution function:

47

For the sake of normalization, the Planck function also appears in the denominator. The Planck distribution is the black body radiation distribution at a certain temperature. The emissivity is calculated by using the Multi-Rad program. Multi-Rad uses a semi-empirical model for calculating the emissivity. The spectral analysis is performed over a wavelength range from 0.4 Am to 20 Amp. t is averaged over incident angles from 0 to 90. The increment in the calculation is = 0.1 gm. The same approach is used for weighting the absorbion coefficient:

The only difference between ,, and ,, is the value of the temperature. For the emissivity, the temperature of the wafer is used, whereas for the absorption coefficient, the temperature of the heater,
IF,

is used. Depending on the way of heating, the

temperature for the absorbion coefficient may differ for front side and backside. n the effective emissivity, no horizontal temperature gradient is assumed, i.e., front side and backside are assumed to be equal in temperature.

48 The one dimensional model used in the literature [ 17] includes the thermal conductivity. The model can be derived by using the heat balance equation:

Under steady-state, the time dependent term vanishes and Equation 2.49 becomes:

Solving this differential equation leads to the lateral thermal diffusion length:

This model can be ephpanded in two dimensions. Then, all space dependent variables are a function of x and y. The pattern shown in Figure 2.12 is represented by two-dimensional periodicity of spacing L. The square pattern can thus be defined by a two-dimensional Fourier-series for a symmetrical square wave:

The coordinates are ph and y aligned parallel to the square edges and the origin is in the square center. This equation is used to ephpress the emissivity and

49 absorbivity in terms of x and y. An unknown temperature distribution can be represented by a two-dimensional Fourier series with coefficients, ti,.

and Equation 2.55 is then solved for the unknown Fourier coefficients t. The ephact solution is given in Appendiph A. The solution is simplified by considering only high symmetry points in the pattern and relating the temperatures at these points to the temperature at the reference point, i.e., the origin.

50 The reference point in Figure 2.14 is in the center of the square. The temperatures at the symmetry points, determined in Equation 2.54, can be ephpressed in terms of cosine functions that correspond to -1, 0, or 1. Table 2.3 gives the cosine values and ephpressions for T(x,y) at the points in Figure 2.14. Table 2.3 Values of Cosines and at High Symmetry Points

51 Phis limit implies that the pattern is decomposed in two different wafers with infinite ephtension. Wafer 1 is like the field and wafer 2 is like the square. By determining S, ephperimentally, one can obtain a result for 2.44.

CHAPTER 3 SPECIFIC PROCESS MODES AND EXPERIMENTS

The heating process in DTP follows a certain recipe for the temperature versus time profile. Process modes differ according to the way in which the intensity is controlled. One method is just to supply a certain power level to the lamps without controlling the temperature directly. This is known as the open-loop intensity control (OLIC). If the temperature is controlled by an error feedback, it is referred to as closed loop temperature control heating. The objective of closed loop temperature control (CELT) is a more accurate temperature control approach, in contrast to the COLIC.

3.1 Open-Loop Intensity Control (COLIC) In open loop intensity control, the wafer is placed into the chamber and the lamps are supplied with constant power. The power control of COLIC operates under the assumption that if the same power is delivered to the lamps, then the same intensity will be emitted. Thus, if the wafer and process chamber conditions do not vary, then the absorbion of the intensity by the wafer will be constant. The issues with power control are chamber window and wall hazing altering the absorbion and reflection property of the wafer environment, lamp aging, wafer variations, and line voltage regulation. The COLIC was tested on wafers with various backside films and an RTO process on the front side. Temperature distributions are inferred in this process by the ophide film thickness. With a constant recipe, it is assumed that there is a relation between power input, temperature, and ophide thickness. Beforehand calculations were done to find out

52

the power requirement for these conditions. The tacit assumbion is that the absorbion, emissivity and reflectivity are the same for a layered wafer as it is for the bare silicon wafer used for steady-state temperature calibration. The chamber was preheated to try to give all wafers the same initial condition, but the first wafer was always a dummy wafer. Temperatures during RTO processing were measured by pyrometers and thermocouples. Film thicknesses on the backside of the wafers were determined from separate monitor wafers. The wafers under test were p-bulk from various slots, ephcept for one which was an epitaphial wafer. Bare wafers from several batches were included for reference points. Table 3.1 gives an overview of the wafers used in the experiments. Table 3.1 Wafers Backside Films Used in the Experiment

54

The polysilicon layers were imdoped. They were annealed in a furnace after CVD deposition. The ophide films include an underlying furnace oxide and CVD-PE-TEOS deposited films that are DATA denazified. Some nitride films were similarly deposited and annealed. Initially, the wafers were cleaned by an HF-RCA clean, which leaves 5 to 7 of chemical ophide on the wafers. RCA clean with HF dip is a standard pre-flimace cleaning step to avoid diffusion of contaminants into the wafer. The hydrofluoric acid (HF) preferentially etches the silicon diophide, but not the silicon. It also removes organics, particles and some metals. The wafers are first immersed in a hot mixture of ammonium hydroxide (4), hydrogen peroxide (1-1202) and demonized (DID) water in a ratio of 1:1:5. This removes traces of organic residues and when done on exposed silicon surfaces, traps ionic contaminants such as sodium and potassium in a thin ophide. The wafers are then rinsed in a quick dump rinser (QDR) which provides a fast method of rinsing, allowing wafers to be moved quickly from one step to the nepht. Next, the thin layer of silicon diophide (5i02) is removed by dipping in dilute hydrofluoric acid (HF). Another QDR cycle follows. Then the wafers are immersed in a mixture of hot hydrochloric acid (HC1), 202, and DI water. The ratio is again 1:1:5. This mixture is effective for removal of ionic contaminants, especially metals. The wafers are again rinsed in the QDR. A final HF and QDR step is obional. Then the wafers are spun dry and are ready for the lamp RTO process [18]. In the open loop intensity control recipe, the recipe time per wafer was 15.7 min. At the beginning of the process, the wafer was placed into the DTP chamber. t was placed in such a way that the layers were on the backside and bare silicon was on the

55

front side. The wafer was ephposed to 20 % 2 for 600 s. Then the 2 concentration was decreased to 10 %, and the 02 concentration was increased from 0 % to 8 % for 10 s. n the third step, with duration of 20 s, the 2 concentration was reduced to 0 and the 02 concentration was decreased to 5 %. n the last step, before the heating step was implemented, the wafer was ephposed to an 02 concentration of 3 % for 30 s. The lamps were then turned on to an intensity of 42.5 % with a rate of 10 %/s. After reaching an intensity of 42.5 %, the lamp intensity was keb constant for 60 s. In this 60 s, the 02 concentration was kept at 3 %. The supplied intensity of 42.5 % for 60 s heated the wafer to about 1090 C. The lamps were then switched off. After 10 s, the 02 supply was also switched off and the 2 concentration was increased from 0 to 5 %. n the last step, the
2

concentration was further raised to 20 % for duration of 220 s. A ripple pyrometer was used for temperature measurements. The ripple pyrometer

compensates for unknown variable emissivity of wafers and removes interference from reflected radiation from the heating lamp. n other words, it separates the wafer remittance radiation from reflected lamp radiation and simultaneously finds the wafer emissivity in situ. Power to the heating lamps is modulated with oscillatory functions of time. Therefore, two separate input obics, one for wafer radiation and the other as a reference for the lamp radiation is used. The electrical signals are separated into a slowly varying quasi-steady and a rapidly varying ripple component. The slowly varying components represent the radiation from the wafer. The ripple signals are digitally processed to find the component of the lamp radiation that is reflected by the wafer under the assumption of negligible wafer temperature oscillation at the ripple frequencies. The ripple signal analysis yields a

56
measure of the wafer reflectivity that is used to compute an effective wafer emissivity. It also allows subtracting Camp interference to yield a more slowly varying signal for the thermal radiance of the wafer. The analysis interval was from 45 s to 55 s. Table 3.2 gives the temperatures measured by direct thermocouple (DTI), the emissivity uncompensated pyrometer (GYRO) and the ripple pyrometer (LPL), i.e., the temperature during the process. All values of temperature are averages over the time interval from 45 s to 55 s measured from the time the lamps are turned on.

Table 3.2 Temperatures Measured with Thermocouple, Pyrometer and Ripple Pyrometer

After the process was complete, the ophide thickness was measured with the Thermawave at 52 points. An area corresponding to 5 mm on the edges of the wafer was excluded from the measurements. Table 3.3 summarizes the measured oxide thicknesses.

57 Table 3.3 Thermawave Measurements of Oxide Thicknesses

Iomparison of the temperature measured by the ripple pyrometer in Table 3.2 with the ophide thicknesses in Table 3.3 shows correlations where the oxide thickness tends to increase with the temperature. For ephample, wafers in the J2 series were heated to a temperature near 1025 C, which is at the low end of the temperature range, and corresponds to an ophide thickness of 45 A, which is at the low end of the oxide thickness range. Other wafers were heated to higher temperatures and the oxide thicknesses were greater but the correlation between the two is not a perfect one. These results show that OLIC does not yield a wafer independent process for RTO. This is a strong motivating factor for selecting closed loop temperature control, which is treated in the nepht section.

58 3.2 Closed Loop Temperature Control (CELT) f coatings on the wafer are present, then the coated regions will ephperience different temperature-time cycles during processing than the uncoated regions, leading to a significant impact on process uniformity. DTP systems are usually operated in the temperature mode by performing a closed loop temperature control by a calibrated obical pyrometer or thermocouple. This can be achieved with a pyrometer that measures the wafer temperature and provides feedback to the control system, such as shown in Figure 3.1 [8]. The challenge is to perform highly accurate pyrometer measurements on a wafer of unknown spectral emissivity in surroundings in which a significant amount of stray radiation may be emitted by the heat source.

Although the incident intensity differs from the absorbed intensity, only the absorbed intensity influences the closed loop temperature control to match the actual temperature-time profile to the desire one. Therefore, the absorbion characteristics have to be known. For the purpose of the calculations, an ideally controlled temperature is

59

assumed. The temperature at the measured spot is assumed to be given at any time by the pre-set temperature time cycle. A signal from an appropriate thermal sensor, thermocouple or pyrometer controls the feedback loop and determines the power to the lamps [8]. In this thesis, all RTP systems that were used to test the model were operated with model based temperature control. Model based temperature control needs to be tuned through temperature data logs of the actual temperature in the chamber during an entire process cycle. Figure 3.2 [8] shows a model based temperature control system with a thermocouple to measure the wafer temperature. Model based temperature control is based on systematic use of a mathematical model of the process in determining control algorithms for the process.

Another possibility is to use a proportional integral differential (PID) control. Precision wafer temperature estimation and control is more critical in RTP due to the lag of the thermocouple relative to wafer temperature under fast ramping conditions. A

60 proportional integral differential (PIED) controller can overshoot the desired temperature and needs more time to tune due to the lag time. Model based temperature has the ability to predict the discrepancy between these two parameters and allow to accurately control wafer temperature during dynamic conditions. The control design relies on the ephperimental data obtained from the reactor to be controlled. The system closed-loop behavior is dependent on both the hardware design and the controller design. The advantage of the closed-loop simulation is that it allows for quantifying the response to a potential disturbance. Decreasing the feedback gain can lead to sluggish response resulting in unaccebable temperature non-uniformities and tracking error. Increasing the gain can create power demands that ephceed the range of actuation resulting in power saturation and loss of control. To evaluate lateral temperature non-uniformities during RTP, two indirect measurement techniques were used in these studies: Annealing of a high dose, low energy As-implanted (wafer maps of sheet resistance) Rapid thermal ophidation The sheet resistance of implanted Si upon RRTA can be correlated with wafer temperature using a calibration curve. Hence, wafer maps of sheet resistance can be interpreted as lateral temperature measurements on the wafer. The sheet resistance depends on activated departs and the activation process is temperature dependent. Hence, partial activation of ion implanted species is very suitable to measure and obimize the across wafer uniformity. The data were converted into effective temperatures using a linear model for dependence of mean post-RTA sheet resistance on temperature. A common disadvantage of both indirect methods is that they integrate all variations from

61 the defined processing recipe into a single parameter. Even if the final temperature distribution across the wafer is accebable, there could be a part of the processing interval in which temperature non-uniformity is above the limit of acceptance. The model uses four temperature states: lamp filament temperature, quartz temperature, pyrometer temperature and TC temperature. A set of four first order, nonlinear differential equations describe the model. Feedback is achieved by updating the model, based on a comparison between actual (measured) system and modeled system response. The Model Based Control uses two steps to reach a tight match between the set point temperature and wafer temperature. In the first step, the 4 state model of the system is inverted to predict that level of power input or position that will drive the wafer temperature to reach the set point temperature. The inverted model transforms Si temperature into predicted lamp voltage or position. In most cases, the calculated voltage or position is limited between a maphimum and minimum value. Due to this limitation, a perfect match is not always possible. This leads to a limit on the first and second order deviations of temperature. In the second step, the actual lamp voltage or position (after clipping) is used as input of the forward model to calculate a prediction of the indicated TIC temperature. Any offset between the predicted TC temperature and the measured TC temperature is used to adjust the model. The forward model is used to calculate the TC temperature (not the Si temperature). In general, any parameter of the model (in the model data base) could be tuned in real time while the control is active. An obimization method can be used to tune the chamber efficiency, which is dependent mainly on wafer absorbivity, during temperature ramps and to tune the wafer emissivity during steady state [8].

62 3.3 Model Validation Experiments

The model described in Section 2.5 was tested by annealing wafers with square patterns. Temperature profile was inferred from the sheet resistance of an As-implanted wafer. The process steps were: 1) Clean wafers and grow pad ophide 2) Implant As at a 20 kV and 10 16 cm 2 3) Deposit 110 nm Si02 4) Deposit 100 Om polycrystalline silicon 5) Pattern squares etched by removing polycrystalline silicon 6) RTA process 7) Strip off pattern films

8) Measure sheet resistance

The ARTA process was a spike anneal with a temperature versus time profile of the type idealized and shown in Figure 3.3 [8] with constant ramping rates Up and ado, and steady temperature
Amax

at the peak. The definition of spike assumes that this is achieved

by heating as fast as possible to the desired temperature and immediately cooling as fast as possible with minimal dwell time at peak temperature. In practice, the actual temperature profile is shown for the furnace process in Figure 2.5 and for the lamp process in Figure 2.9. In modeling the spike anneal, the process temperature is taken to be
Amax.

63

r figure 3.3 idealized se process [].

In this work, the spike process was realized in furnace and lamp RTP systems with three different kinds of wafers. In the lamp DTP system, the power input is controlled. In the furnace system, the wafer position in the furnace is controlled. As discussed previously, in the furnace system, the power is kept constant, i.e., the furnace is heated steadily. A temperature gradient is realized within the furnace and the wafers are elevated in the furnace. The temperature in a furnace is a function of the position. The wafers under test were bulk wafers, epitaphial wafers and SOl wafers (SIMOX), as illustrated in Figure 3.4.

64

Figure 3.4 Cross sections of a) bulk, b) epitaphiall, and c) SOI wafer.

65
Each of these wafers were coated with Si02 and polysilicon layer. For realizing the pattern, the polysilicon layer was etched away in a square pattern. The Figure 3.5 shows the schematics of the wafers after they were etched. Figure 3.6 is a full wafer view of the front side pattern, which is an array of 1 cm ph 1 cm squares, spaced L = 2 cm apart.

Figure 3.5 Wafers after the etch process.

The measured thicknesses of the layered structures are shown in Tables 3.4 and 3.5:

Table 3.4 Measured Specification of Layers on Bulk, Epitaxial and SOIL Wafers

Table 3.5 Measured Specifications of Bulk, Epitaphial and SOIL Wafers

(R

The wafers were placed in the chamber in a way such that the pattern was on the front side. In the case of the lamp system, only the front side was heated, whereas in the furnace, both sides were heated. Different measurement methods for wafer temperature were used. Multi-Dad was used to calculate the spectral emissivities of the wafer at
Amax

= 1050 C corresponding to the DATA process. This quantity is multiplied by the Planck function. In Figures 3.7, 3.8 and 3.9, the remittance are plotted as function of wavelength for bulk, epitaxial and SOIL wafers at room temperature, 600 C and 1000 C, respectively.

69

70

temperatures.

After integrating and normalizing these curves, the emissivity and absorptivities shown in Tables 3.6 to 3.9, were calculated. For the emissivities, the temperature refers to the wafer temperature. For the absorptivities, the temperature refers to the heater temperatures. Table 3.6 Calculated Emissivity at 1050 C

71

Calculations were performed at a wafer temperature of 1050 C. The wavelength range was between 0.4 and 20 gm with an increment = 0.1 gm. The values calculated with Multi-Rad are all hemispherical emissivity. As can be seen from the Table 3.7, the backside emissivity of the bulk wafer and the SOl-wafer are equal. Both wafers do not have any layers on the backside, whereas the epitaphial wafer has a polysilicon/Si02-layer. At the front side, the bulk and epitaxial wafer have a similar emissivity. The SOIL wafers differ in greater magnitude in emissivity, especially for the case in which the polysilicon is etched away. It can be inferred that the buried ophide influences the emissivity. The values of emissivity for the furnace process were calculated as follows. In the furnace process, the heating of the wafer is from both sides. For the furnace, a

72

In the furnace heating, a reflection coefficient for the upper part of the chamber of 0.05 and for the lower part of the chamber of 0.15 is defined. The results for the effective emissivity were obtained by utilizing Equations 2.44 and 2.45 and are shown in Table 3.10.

In a lamp process, the wafer is only heated from the front side with a temperature of 2650 K. Further, it is assumed that the wafer temperature on the front side and backside are the same. chamber reflection coefficient of 0.504 was used. The calculated effective emissivity and absorbivity are shown in Tables 3.13 and 3.14.

73 Table 3.13 Calculated Effective Emissivity for Lama Heating at 1050 C

These values can be used in Equation 2.43 to calculate the ratio of temperature 1 to T2. This ratio can be compared with the experiment and used for determining the Fourier coefficients discussed in Section 2.4 and in Appendix A.

CHAPTER 4 RESULTS AND DISCUSSION

Two experiments were conducted, using DTP with open loop on patterned wafers and closed loop on patterned wafers. The open loop experiment, in which the lamp power was controlled, shows that emissivity variations among the wafers produce process and temperature variations. The second experiment used patterned wafers in which temperature was maintained under closed loop control. The model for the effect of emissivity patterns on wafer temperature, presented in Section 2, was found to give a good agreement with the experiment described in Section 3. Tables 4.1 and 4.2 show the results obtained from the model and from the closed loop temperature control experiment, respectively. The T 1 /T 2 is the ratio between the absolute temperature at the origin, i.e. the square, and that at the point 4 in are in Kelvin).

Table 4.1 Model Calculated Temperature Ratios

74

75 Table 4.2 Temperature Ratios based on Experiments

Comparing Tables 4.1 with 4.2, a good agreement between the model and the experiment can be noted. value for T, /T2 of unity can be interpreted as absolute uniform temperature distribution across the wafer and thus, a uniform process for the specific wafer. In all cases, 1 <
T2,

since square areas 1 with the oxide film heat more

efficiently than the field areas 2 with the poly/oxide film. This is quantitatively suggested by inspection of the emissivities and absorptivities shown in Tables 3.10 3.14. It is observed that the epitaxial wafers and the bulk wafers produce similar temperature non-uniformity across the wafer. In the furnace process, the experiment gives a temperature non-uniformity of T 1 /T2 = 0.987 for bulk and epitaxial wafer, whereas for the model, the calculated values are 0.977 and 0.978, respectively. The lamp process shows for both, bulk and epitaxial wafer, more temperature non-uniformity across the wafer than for the furnace process. This can be seen by comparing T1/T2 0.8 in the iarnp system for both wafers with 0.9 in the furnace system for both wafers. It can be inferred that the furnace process is more uniform for bulk and epitaxial wafers.

However, it has to be considered that the lamp system is a single side heater, whereas, in the furnace, the heat radiation is incident on both sides. For the SOIL wafers, both the processes, furnace and lamp, produce comparable results. These wafers also show a good agreement in /T2 between the model (0.959 for furnace and 0.966 for lamp) and the experiment (0.948 for furnace and 0.946 for lamp). It should be recognized that the obical property of polysilicon and epitaphial silicon were approximated within the model. For polysilicon and epitaphial silicon, the property of crystalline silicon were used in calculating emissivity with Multi-Rad. Another simplification relates to neglecting temperature dependence of optical property of materials other than silicon. For Si02 for example, the n and k functions are taken to be temperature independent. The calculations in Multi-Rad can be very time consuming, approximately 30 minutes per wafer, and the program produces interpolation errors. In the model, the fields and squares do not interact with one another, i.e., field and square are treated as separate wafers. In reality, the emission from field gets reflected in the chamber walls and therefore, influences the square, as well

as vice versa. A further

improvement of the model would be to include a different reflectivity and/or calculate the reflectivity of the wafer by using Multi-Rad. Further, the wafer is considered as

an

infinite plane. The higher emissivity at the edges, which also influences the temperature uniformity, is neglected. For DTP systems, steady state is not always a given. As the processing time gets shorter and the thermal budget is reduced, the steady state is less possible. A further improvement of the model would be to include the transient heating and cooling. The model is only a two-dimensional model, even though heat diffuses in

three dimensions. Since the wafer is very thin, the steady state approximation is justifiable in the present case. perfectly uniform temperature profile is achieved only when, for all parts of a wafer, the primary radiation flux and the reabsorbed heat radiation are balanced. Several attembs to solve this problem are given in the literature. synergy between the temperature distribution and energy loss is necessary. One of the problems in obtaining uniform temperature profiles across a wafer surface originates from the fact that one starts with a nonuniform primary radiant flux. Even in the ideal case, the radiation from the wafer is not uniform, because of center-to-edge radiative differences and local differences in absorbion and emissivity induced by patterns on the wafers. radiation back onto the wafer enhances these differences [19]. It is evident that a reflective chamber reduces the effective energy loss, and thus the pattern-induced non-uniformity. By changing the method of illumination, the effect of patterns on the temperature non-uniformity can either be amplified or reduced, e.g. double side heating or single sided heating. To suppress non-uniformities during steady state, it is best to illuminate only the unpatterned side during the ramping stage and to compensate the energy loss during stationary state by double-sided heating. Dynamic non-uniformities can be reduced by increasing the contribution of backside heating with respect to front side heating. Here it is clear that transient non-uniformities is also suppressed by increasing ramp-up times [20]. Some DTP chambers also rotate the wafer in order to get a uniform temperature distribution. Modeling a rotating wafer will be more challenging since there is a time dependence.

78
Experimental results from the literature also demonstrate the possibility of pattern-effect minimization using a hot shielding wafer [20]. Kirsch [21] showed in his paper, that there is no pattern effect when using a susceptor to control the wafer temperature in a single wafer reactor. The wafer was oriented towards the susceptor and the wafer and susceptor had the same temperature. There was a minimal effect because the configuration is not ideal: There is some radiation entering from the side [21]. The influence of the deposition system and patterns on the thickness uniformity can be further reduced by choosing an obimum chemistry. Processes with lower activation energies will result in better process uniformities. Furthermore, it is important to point out that not only the lateral but also the vertical temperature gradient can be a decisive factor. Given the case of front side heating, limited thermal conductivity and interference on the patterned layer at the wafer surface creates a temperature gradient vertically across the wafer. In the experiment, the furnace was a heat source in which the wafer was heated from both sides. This can be compared to the lamp system, i.e., a single sided heating system. The experiments reported in this study show that the temperature difference in the furnace system is smaller than in the lamp system.

APPENDIX A DERIVATION OF THE HEAT BALANCE EQUATION

In this appendix, the derivation of the heat balance equation of a squared patterned wafer in two dimensions is derived by using Fourier analysis. The heat balance equation and the general 3 D Fourier heat conduction equation is used to describe the temperature profile within a heated wafer. The thermal diffusion equation is given as:

In A.1 p is the density of the material, C the specific heat of the solid and q the heat flow. The density is assumed to be homogenous. In a solid, there is negligible difference between the specific heat at a constant volume c,, and a constant pressure c ps . (T) and t , is the thermal conductivity and wafer thickness, respectively. The heat flow in this case is the difference of the heat absorbed (heat source) and the heat emitted (heat loss). For 1 dimension:

In A.2, the variable x is the position, the temperature and t the time. The first term expresses the thermal conduction in the wafer. The thermal conductivity is temperature dependent. (T) is the thermal conductivity and t , the thickness of the wafer. To the thermal conduction, the heat absorbed at the position x is added. The absorbed heat energy density
qabs(x,T)

is obtained using the following equation:

79

80

where, Q(T) is the heat input power and is defined by the Stefan-Boltzmann law and F(x) is the absorption coefficient at the point x. The heat loss is defined through the emissivity. This differential equation can be solved analytically. Taking a closer look at the solution, the thermal diffusion length can be expressed as:

Equation .2 also can be extended to two dimensions. two-dimensional pattern function is superposed with the emissivity and absorptivity. This two-dimensional pattern function is realized with a two-dimensional Fourier square wave series. This series is centered in the square, i.e. it is a symmetrical square wave, with an offset to avoid negative emissivity and absorptivity.

n=1,3 ...

Equation .5 is a one-dimensional square wave Fourier series with the Fourier ['his function is made symmetrical about the origin by shifting x by L . Th i s i ntroduces a phase factor. 2

two-dimensional pattern function is created by multiplying a one dimensional pattern function F(x) with a pattern function Fay).

81

And considering only x and y dependence, the result can be simplified further:

There are five different terms, terms without any Cy dependence, terms with 0 2 T (x, y) , dependence. Therefore, it is possible to combine

As an approach to solve equation A. 10, for Tax,y) the general Fourier series formula is used.

82

83

85

86

87

88

This is the exact expression for the Fourier coefficient. For y, the same result is obtained. These tn can be used in Equation A.11 to calculate .

REFERENCES

1. Fair Richard B. a1993). Rapid Thermal Processing A Justification. Rapid. Thermal Processing Science and Technology. Academic Press, Inc. 2. Hebb J. P., Slaves Jensen a1996) The Effect of Multilayer Patterns on Temperature Uniformity during Rapid Thermal Processing, Journal of the Electrochemical Society. Vol. 143, No. 3, pp. 1142 1151. 3. Meschede D. a2002) Gerthsen Physic. (21 St Edition) Springer Verlag. 4. Kekac S., Yener, Y a1985) Heat Conduction. a2"d Edition), Hemisphere Publishing Corporation.

5. Roozboom Fred. (1993) Manufacturing Equipment Issues in Rapid Thermal. Processing. Rapid Thermal Processing Science and Technology. Academic Press, Inc. 6. Lojek B. a1993) Issues in Manufacturing Equipment Issues in Rapid Thermal Processing. Rapid Thermal Processing Science and Technology. Academic Press, Inc.

7. Kakoschke R., Is there a way to a perfect rapid thermal processing system? Materials Research Society Proceedings, Volume 224, pp. 159-170 a1991). 8. Retrieved August, 2005 from the World Wide Web: http ://www2. Mgr. Ruh. edu/smotamar/RTP/RTP.html. 9. Fiery A. T., Ravindra N. M. a2005) Film Emissivity Patterns in Rapid Thermal Processing of Silicon Wafers. TAMS Symposium 2005.

10. Retrieved from the World Wide Web: http://www.axcelis.com/. 11. Crisham Z., Massed, a1993) Rapid Thermal and Processing of Dielectrics Rapid Thermal Processing Science and Technology. Academic Press, Inc.

12. Retrieved from the World Wide Web: http://www.appliedmaterials.com/. 13. Retrieved August, 2005 from the World Wide Web: http://ece-www.colorado.edu/bart/book/eband5.htm. 14. Ravindra N. M. a2004) Lecture Notes "Physics of Materials", Fall Term 2004. 15. Retrieved August, 2005 from the World Wide Web: http://www.vortec.com/fundamentals.html.

89

90
16. Hebb J. P., Slavs J. F. a1997) Multi-Rad User Manual. MIT, Cambridge. 17. Niess J., Berger R., Titans P. J., Nenyei Z. a2002) Pattern Effects and How to Explore Them. 10 th International Conference on Advanced Thermal Processing of Semiconductors. DTP 2002. 18. Retrieved August, 2005 from the World Wide Web: http://www.rpi.edu/dept/cieem/mcrusers/tools/wetbench/www/Wafer_Cleaning_ Procedures.htm. 19. Vandenabeele P., Manx S., De Keersmaecker R. a1989) Impact of Patterned Layers on Temperature Non-uniformity during Rapid Thermal Processing for VLSIApplication. Mat. Res. Sock. Sump. Vol. 146., pp. 149 160. 20. Nguyen L. H., Diet W., Niess J., Nenyei Z. Logic in RTP. IEEE RTP'99 Conference Proceedings, pp. 26-38 a1999). 21. KirschA.a197)BenftdLimaosRtvelyHdSuscpor. Rapid Thermal and Integrated Processing VI. Materials Research Society Symposium Proceedings Vol. 470 pp. 159 173.