Lec 46

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Digital VLSI Testing

Prof. Santanu Chattopadhyay


Department of Electronics and EC Engineering
Indian Institute of Technology, Kharagpur

Lecture - 46
System/Network – On – Chip Test (Contd.)

(Refer Slide Time: 00:21)

So, testing principles, so this diagram we are bit familiar, but what is happening is that
the input patterns they are applied to the digital circuit, so the responses are collected, so
they are compared with the stored correct response and get the test result whether it is
true or not; but after this has been done, so we have to screen the defective chips at the
wafer level or at the package level. So, the defective chips are to be separated then the
stress test, so it has to do a, so if the chip is operated for a number of continuous hours
then whether the chip starts malfunctioning or not sorry the stress test or burn in face.

Then for the diagnosis purpose, so we need to locate the defect and then we see what is
the yield that we are getting. So, yield learning actually what happens is that in some in
some production if we find that the chips are failing, they are not passing the tests then
we try to figure out like there may be some problem with the manufacturing process, so
we try to locate the defects. And if a number of chips show the same defect location that
means, there is some problem with the manufacturing process. So, the yield got
decreased, so we need to correct those process, so that the yield will increase again.
Then speed binning, so there may be different versions of the chip like some of the chips
they may not be able to operate at a very high frequency, they may be able to operate in
the kilo hertz range, but not in the giga hertz range. So, even after the this fabrication
process everything is similar, so due to some process variation, so it may so happen that
we cannot reach some frequency level of operation for some chips. So, that way we may
divide the chips into different speed bins and then that you can say that some of them are
slow version some of them are fast version like that.

Then design for testability, so that is typically used, so that we can do this testing process
in a methodical fashion. And then this test pattern application response gathering
analysis they can be done in a very well defined way. Test generation to generate the test
pattern and scan design, so these are the various issues that we have in the testing
process.

(Refer Slide Time: 02:46)

So, why do we go for this core based SOC testing, why is it so important. So, you see
that system on chip integrated circuits based on embedded intellectual property cores are
now commonplace, so this is a very now there are many components many chips that are
being designed which are designed as SOCs. In fact, the embedded VLSI system has
become very popular and then it is actually this SOC based system design.

SOCs include processors, memories, peripheral devices, IP cores, analog cores all that,
so any company who is involved in digital IC design analog IC design, so they are also
designing the cores. Because as a single chip solution, so that their design has to go
inside those systems corresponding to some application, so we need the core based
design. So, this SOC becoming more and more popular, so this core vendors or IP core
designers they are also becoming more and more important.

So, for example, this arm processor, so arm is never designed as a chip, so it is always
designed as a core. So, depending upon your requirement, so we can just take some part
of arm processor whereas, we do not take other features of the arm. For example, the
floating point portion may not be required for my operation, so I do not take the floating
point part in it. I may or may not be interested in the security aspect, so as a result I may
or may not include the security part in the processor design. So, the processor itself is
getting designed as a core only, so that depending upon the requirement, so I can
customize and I can take only that core into my system. So, this SOC design, now
include this core based designs.

Low cost, fast time-to-market, high performance, low power. So, low cost because we
are not wasting time in getting the chip fabricated for this individual components. And
since the vendors, so they are providing a net-list which is they are providing a layout
level description, so that is at a much higher-level, so that is they did not have to go into
this fabrication process of the chip, so this cost of this core is going to be much less than
the cost of the chip.

Then fast time-to-market because now as a designer, so if I think that my system will
have a number of different components like say CPU, memory, DSP etcetera and for the
sake of best performance what I need to do is to take all of them into a single chip. Now
if I myself try to design all these components or in a then it will take huge amount of
time; and no company that we have, so it is it is possible that we have got expertise in all
these domains. So, every company they do specialize in one or two domains, so that way
this if I go one for high performance may be some company is doing high performance
CPU design, somebody is doing high performance DSP processor design, so that way
there may be issues. So, that way I need to take their designs into my system, so that is
why we have to go for this core based SOC design.

Low power, so power minimization is again another very important issue. And again this
IP cores that are getting generated that are getting developed, so this power may be a part
there and this they may be optimized better for the power, so that way I have to take I
can go for this cores which are branded as low power versions, so they will consume less
power.

Manufacturing test needed to detect manufacturing defects. So, this is very important
because now previously as in a board level design this individual chip testing was not
that much important, but here all the cores are to be tested. And in fact, this particular
road map it shows how this manufacturing cost has in increased. So, over the years, so if
you look into this number of transistors cost per transistor in terms of cents, so it has
reduced drastically. Why, because the large numbers of transistors are being fabricated
on the same die, so cost part transistor it has come down significantly.

On the other hand, the test cost though it initially dropped with the increase in the
number of transistors, so the ratio actually made it to reduce in terms of this ATE cost
and this cost per transistor, so that way it was reducing. But now it has again started
increasing, so from see 2000 onwards, we can see a steady increase in the test cost. And
it is very much likely that there will be a there will be an overshoot that test cost will
cross the manufacturing cost and that is because these ATEs are becoming costly, so if
we want to test more and more chips at high frequency. So, if we want to test that ATE
cost are going to be high, so that way the test cost is going to be high. The complexity
the amount of test time to be devoted is going to be high, so it is going to have some
good amount of time requirement, which is translated into the cost of the system, so test
cost is going to overcome the manufacturing cost.
(Refer Slide Time: 08:35)

So, this is a typical system on chip architecture, you see that we have got a number of
core CPU core, memory array, legacy core, IP hard core then this self test control, DSP
core, interface control, embedded DRAM etcetera. There may be some so we have got
some I O pads around it through which this I O pins for the chip are available. For serial
testing boundary scan structure, this 1149.1 tap controller is introduced, so that I can put
a boundary scan cells around it and that way it can be operated that it can be tested.

Now, if you look into this system you see that the test access is limited, so for accessing
this DSP core, so only may be only a few of the pins are available on the I O pads, rest
are not. And in the worst case, it may so happen for that some of the cores which are
embedded deep inside the system none of the pins are available at the system I O pads.
So, then how do we apply the test patterns to them, how do we see what is the response
generated by them. So what is this, so test access is limited, so what is required that test
sets must be transported to embedded logic. So, as we are discussing in this example that
this test pattern must be transported from the test at the source that may be an ATE may
be a BIST, whatever it is from there through this TAM it needs to be transported to this
circuit under test the core under test and the response is to the sink.

And third important point is high test data volume and test time. So, high test data
volume because large number of patterns for every core and there are a large number of
cores, so total test data becomes very large. So, can we store, so much of test data in the
ATE memory, so that becomes a fundamental question. If the answer is no, then we have
to use some sort of test compression mechanism, test vector compression mechanism.
And then as soon as we do that then there is a decoding stage that will be coming, so
before applying the coded patterns to the core for testing we have to decode them and get
back the original pattern set. So, that way it is going to happen that this test time and test
data volume they are going to be an important factor.

So, even if the ATE allows you to store a large volume of data test data in it test large
number of test patterns in it, so you need to transport those test patterns from the source
to the core under test. So that itself will take a number of clock cycles number of ATE
cycles to transfer over the access mechanism, so that way the test time will become high.
So, high test data volume has got two problem one is the ATE memory requirement will
be high, and the second point is the test time for the chip, so that will also become very
high.

So, this is a typical example like NXP Nexperia processor PNX 8550 SOC, so it has got,
so many flip flops 274 embedded cores, 10 million logic gates and 40 million logic
transistors, so that is a huge amount of space huge amount of complexity that we have.
And now we must be able to test such complex systems in a SOC paradigm.

(Refer Slide Time: 12:12)

Cost of test: The emergence of more advanced IC s and SOC semiconductor devices is
causing test cost to escalate to as much as 50 percent of the total manufacturing cost. So,
this was a statement in 2002, so now it is 2016, so naturally the cost has really become
much higher 70 almost 60 to 70 percent of the manufacturing the cost the test cost has
gone up. As a result semiconductor test cost continues to increase in spite of introduction
of DFT and it can account for up to 25 to 50 percent of total manufacturing cost. So, a
test cost becomes a huge part o though we have got DFT scan and all that, so they are
going to make this testing process easy, but the sheer complexity or sheer volume of test
itself is making the process very costly.

And test may account for more than seventy percent of the total manufacturing cost. So,
this is that I have already said. And test cost does not directly scale with transistor count,
die size, device pin count or process technology. So, this is another problem. So, what
happens is that, so for a VLSI manufacturing process, so as you are increasing the
number of chips that you are producing, so the per unit cost comes down. So, we have
for VLSI chip design and manufacture, so we have got two cost, one is the non recording
engineering cost which is the design part of it, and then per unit we have got some cost.
And since we are doing this in a massively parallel way the large number of chips are
manufactured simultaneously, so this non recording engineering cost is high, but this unit
cost is low, so that way overall cost comes down while we manufacture them in large
volume, so that has happened. So, the designers they are happy with producing more and
more number of chips with the same design.

However each of these chips that are being produced are to be tested, so you cannot
leave one chip untested. So, this increase in the number of chips that are being produced,
so also have proportionately increases the test type, so test cost will go up, it does not
scale down or rather scales up with the number of chips that we are going to fabricate
that we are going to manufacture. So, we can say in some other sense that with the
increasing transistor count, so test cost also increases with the number of dies, it is
increases, number of device pin count it is increasing; that means we have got more
functionality to be tested. And with the process technology, we are putting more and
more chips on fabricated simultaneously, so the cost is also increases.
(Refer Slide Time: 15:16)

So, what is required is some sort of modular testing. So, test embedded cores using
patterns provided by core vendors, so that is definitely there, so we cannot avoid this
situation. And test access mechanisms needed for test data transport. So, what is
happening is that, so this is the ATE. From ATE, it has to transfer data to the embedded
core via some test access mechanism or TAM, and again the responses are to be
collected through some TAM and it has to be given back to the ATE and then this ATE
will again analyze it. So, test access mechanisms needed to for test data transport, so
TAMs will impact test time and test cost.

So, if you can have a wider TAM then you can transfer more test bits parallely as a result
test time will be less, but wider TAM means it will be requiring more amount of silicon
area. So, as a result and that TAM has to be routed, TAM is not feeding a single core, so
it is feeding a number of cores, so that TAM also needs to be routed through the chip. So,
as a result this becomes a very important constraint. So, the test cost will increase
because I have to provide more amount of the silicon, more amount of footprint on to the
silicon for the TAM or holding the TAM.

Then this test wrappers they translate test data supplied by TAMs, so TAM is having the
test access mechanism or TAM, so it has got a fixed width say 32 bit width. Now, each
of these cores may not have 32 input some of them may have say 100 input, some of
them may have say 64 input, some of them may have 90 input like that, so we need to do
some sort of adjustment between the TAM lines and the number of inputs and outputs for
the cores. So, that is how this test wrappers will take care, so it will translate this test data
that is supplied by TAM into a format by in which you can apply the pattern to the
individual components in the core.

This TAM optimization test scheduling, test compression these are very critical issues
and test data volume and test testing time in 2010 was expected to be 30 x of that of
today’s chips that was predicted in ITRS 2005, so that is that is crossed now. So, there is
a huge increase in the test data volume and the testing time, so that is the issue, so we
have to handle such complex situation.

(Refer Slide Time: 18:01)

So, we have to know some sort of test planning. So, we just cannot leave it to the adhoc
solution, so we have to do some sort of test planning. So, we need to optimize test access
to cores and we also need to do scheduling of test hardware. So, how this is done, so we
have to do some test hardware planning, so test hardware planning, so cores are to be
taken, so cores are for a particular design, so cores are imported, the cores are integrated,
so how the cores will be connected between each other, so that is basically the designer
part. Once that has been done each of these cores they need to be test wrapped, so we
have to put the we have to design the corresponding test wrappers for each of these
cores, and then we have to design the TAM like how these cores will be accessed in a
test access mechanism. So, how they are going to be utilized in a test access mechanism,
so that is the TAM design has to be done.

And after that we have got this top level DFT, so that will have this test control blocks
IEEE 1149.1, so they are to be integrated. So, this is the thing. So, you see that it starts
much early in the manufacturing process, so after the cores are been imported and they
are been integrated into the design, so we know what are the cores, so what are their test
requirements and all that, and immediately we have to start doing the other operation that
is wrapper design, TAM design etcetera. And on the software side, so once the core has
being imported side by side we also get the corresponding test data. So, test patterns and
the correct responses, so they are also provided by the core vendor, so they have given
the core as well as the test part test pattern part.

And then at atop level ATPG can be done, so that we can generate the test patterns for
the glue logic. And basically the glue logic are the portions which are extra small logic
that are added for making the functionality of the chip complete may be this requires
some translation of data from one format to another format or some extra small, small
logics to be added inside the design to make the design complete. So that is the glue
logic. But for testing this glue logic, we have to apply test pattern, so to know what is the
test pattern for testing this glue logic, so we have to take help of some ATPG.

And also sometimes, some cores are available as soft cores. Soft core means the design is
provided at the behavior level itself or may be at most, yeah it is provided at the behavior
level. So, if it is that then we can generate test patterns at that point, so it is not at a
layout level, but at a much higher level which at the level of may be at the level of net-
list may be at the level of behavior. So, I can run some ATPG tool to generate the test
vectors for them, so this for soft cores also, we can do it. But for hard cores we cannot do
anything for hard cores we have to rely on the patterns provided by the core vendor. So,
these test wrappers are to be designed.

And then we have to go for the scheduling policy. So, we have to find out like how when
are you going to schedule which core for testing, and we have several constant like say
this wrapper this TAM should be available to which this core is attached. Then the power
limit should not be violated the thermal power should not be violated then there may be
some precedence constant, so keeping all these things into mind, so we have to come up
with a test schedule. So, that gives rise to a number of alternative works that are been
reported in the literature on this test scheduling algorithms. And ultimately the test
assembly; test assembly means that I have got this scheduling algorithm, I have got the
compression mechanism, I have got the response analysis mechanism, so all these things
are to be integrated together for the whole system to operate, so that is the test assembly
process. So, these are the two important points, the wrapper design TAM design and test
scheduling these are the two important points to be addressed in test planning.

(Refer Slide Time: 22:31)

So, 1500 core test standard, so this is provided by IEEE. It defines test interface between
core and SOC, provides core isolation and plug and play protocols, so these are the
goals. And the scope is to standardize core isolation protocols and test modes, but so this
tells like how can we make a core isolated from others, and apply the test pattern to that
core only. So, we have seen previously that rest of the cores put into bypass mode and
whenever we are transferring test pattern for a particular core, so we put others in bypass
mode, and then that test pattern is applied.

So, either serially or parallely whatever we can think about, but the rest of the thing like
the TAM design type of test to be applied test scheduling, so these are not answered in
the 1500 standard. Because this is going to be this is TAM part may be the parallel
access, so that is an optional part of 1500, so though the instructions and every things are
provided, but it does not tell how to design a tam, so that is left to the designer. So, the
test engineer may come up with different TAM architecture and accordingly after the
TAM architecture has been designed can decide upon the test scheduling problem how to
schedule different tests for the application.

(Refer Slide Time: 24:00)

So, this is that 1500 wrapper, so we have got the normal mode of operation, serial test, 1
to N test, bypass, isolation and extest, so these are the various modes. So, we have
already discussed this thing that this WSI the serial input can be used for giving the serial
data some we may like to configure this boundary scan registers, boundary scan sales
they are through the serial input, and or we can have some parallel input as well. So, this
test stimuli can be applied parallely also this wrapper parallel input and wrapper parallel
output, so using that also this WBRs can be configured. And WBR can also get input
from the functional data that is in the normal mode of operation, so it can get the
functional data and then the functional data can be applied to the core, and they will be
applied functional, so that is the normal operation of the system.

In the test mode, we have got serial tests in which through this WSI line, the test patterns
can be loaded into this WBCs that we have here. Or we can have 1 to N test, so there is a
multicast sort of thing, so one from one point, so I want to test inputs parallely, so all of
them will be loaded simultaneously, all those scan chains will be loaded simultaneously.
Or it may be a bypass mode, where this through this WBY wrapper bypass register, so it
will go to WSO, so this that mode is there.
Then isolation, so we can put into the safe mode basically that safe mode that we had
discussed previously, so that is the isolation mode. And this extest, so if you are trying to
test the interconnect, so you can put this into extest mode and where this output, so we
can load this WBR. And after that whatever interconnects are there this functional data
wherever it connects o the next core, so there we can check whether the value have been
BIST properly or not, so that way we can have some extest mode also we can test this
glue logic part using this extest mode that we have seen previously.

(Refer Slide Time: 26:09)

So, this is the wrapper boundary cell structure, so quickly looking into it. So, from chip
this is the functional input, and we can have this WSI. So, for serial inputs, so we can
have this WSI or WPI, so they will be coming. And this shift bit will be given, so if it is
0, then this WSI, WPI they will be coming to this flip-flop and they will be loaded.
Otherwise, if you want to load some functional value then from the core, so this value
may be taken back and put here. So, we have got this type of boundary input cell and this
is the wrapper boundary output cell. So, in the output cell, so this is going to the WSO or
WPO line, and this is going towards the chip, so that way we can design the output cell.
So, here from the WCI line, so it is controlling whether the input will come from chip or
it will come from this serial this wrapper lines WSI or WPI that will be applied to the
core. I will continue in the next class.

You might also like