4

Cell-Balancing Techniques: Theory and

Implementation
4.1

Introduction

In the safety chapter we briefly discussed the issue that when multiple cells
are connected in series, the cell voltage is not always equal to the pack voltage
divided by the number of cells. How does this happen? This chapter explores
that question in detail, but the first question to answer is this: Why do we care?
The first reason is safety. Remember, when lithium ion cell voltage exceeds 4.2V by a few hundred millivolts, it can undergo thermal runaway, melting the battery pack and device it is powering. It can even blow up as a big ball
of fire. Although a well-designed pack has an overvoltage protection circuit that
will prevent such an event (usually even two independent circuits!), it is better
not to tempt fate by triggering this protection unnecessarily.
The second reason is longevity. If the maximal recommended charging
voltage is exceeded even a little, it will cause very accelerated degradation. Just
increasing the charging voltage from 4.2 to 4.25V causes the degradation rate
to increase by 30%. For this reason, the misbehaving cell that has higher than
its due share of voltage will degrade faster.
The third reason is incomplete charging of the pack. Lets assume the protector circuit does its job and that charging stops when just one cell gets close
to unsafe conditions. Now we have successfully prevented thermal runaway,
but all of the other cells now have lower voltages and are not fully charged. If
we look at the pack voltage, it will be much less than 4.2V multiplied by the

111

112

Battery Power Management for Portable Devices

number of cells. Less pack voltage means less pack energy. (It also usually means
less available capacity, as we will see later.)
The fourth reason is incomplete use of pack energy. Lets consider another
situation. Instead of having too high a voltage, one cell could have too low a
voltage compared to others when the pack is close to the end of discharge. A
pack protector will prevent overdischarge (which would damage the cell) by
stopping the discharge of the whole pack when one cell voltage goes below
the cell undervoltage threshold (usually around 2.7V for a LiCoO2-based cell).
This means that all other cells are still at higher voltages and have energy left.
The pack still has energy, but the device can no longer be used because of one
misbehaving cell.
Now that we have established that cell voltage differences are harmful
enough to take action to remove them, lets look at the causes of these voltage
differences. The first thing to understand is that a voltage difference is not in
itself an imbalance, but a manifestation of the differences in state of charge
(SOC) of the cells if no current is flowing, and of the cell resistance differences
if current is flowing. If we try to instantly eliminate the voltage differences
themselves (e.g., the effect) without eliminating their cause, we will potentially do more harm than good, while wasting hardware resources (cost, size)
by overengineering the balancing circuit to provide huge currents that will be
required for such an instant result, and wasting energy by unnecessarily passing
currents back and forth. Unfortunately, this is exactly what happens in some
commonly used balancing schemes that have been designed without an understanding of the underlying mechanisms of the imbalance. The plan is that
after reading this chapter you will never design or use such an inefficient system
because you will be endowed with a perfectly clear idea of what an imbalance is
and how it can be eliminated in the most theoretically efficient way.

4.2 Types of Battery Cell Imbalance That Affect the Charge/

Discharge Voltage
4.2.1 State-of-Charge (SOC) Imbalance

A SOC difference is the only cause for cell voltage differences if no current is
flowing, known as open circuit voltage (OCV). Indeed, there is a simple correlation between SOC and voltage for any battery chemistry in the form of OCV
= f(SOC,T), where SOC is the state of charge and T is temperature. The form
of the function is different depending on the chemistries, but in general it is
clear that for a given dSOC you get some difference in voltage, dOCV. What
could be causing these differences in cell SOCs? Lets look at a few possible
reasons.

Cell-Balancing Techniques: Theory and Implementation

113

First we consider inaccuracies in the voltage measurement of cell formation cyclers. Most Li-ion cells are cycled after assembly to form a passivating
layer on the anode and to detect abnormal cells. At the end of the cycling, all
cells should end up in the same state of charge (as indicated by the same voltage).
However, cycling equipment is not perfect; there are some channel-to-channel
variations that result in cell SOC differences. To reduce these, of course, the cyclers themselves have to be kept well calibrated, but also, after cycling, cells need
to go be graded, a process in which the cells are grouped based on close voltage,
usually within 2 mV from each other. Now, 2 mV may not sound like much,
but keep in mind that cells are stored and delivered in 50% SOC. It happens
to be close to the flattest portion of the voltage curve, where a 1-mV difference
roughly corresponds to a 1% SOC difference. This same 1% difference in SOC
will result in up to a 10-mV difference by the end of charge and a 100- to 500mV difference by the end of discharge (depending on how deeply the pack is
discharged) because the voltage/SOC curve is much steeper in these areas. Note
that while the percentage of SOC imbalance remains constant during the entire
discharge period, voltage differences among the cells vary with SOC because
dV/dSOC varies with SOC. Figure 4.1 shows OCV differences among cells at
a constant SOC imbalance but at different states of charge.
This figure shows the dependency for a Li-ion cell. The actual shape of the
curve will vary with chemistries, but the concept that SOC differences remain
constant regardless of SOC is still valid. Some chemistries, such as a lead-acid
chemistry, experience large and almost linear changes of voltage with SOC so
it is quite easy to estimate how much dSOC is between the cell for a given dV.
Other chemistries, such as LiFePO4, have an almost completely flat voltage
profile, so even differences in SOC between the cells as large as 5% to 10%
are not noticeable in the cell voltage when it is in the midrange of SOC but
can nevertheless cause drastic voltage deviations close to the end of charge and
discharge, causing protection electronics to trip and to shut down the charging
process before the pack has a change to be fully charged. It is also more difficult to grade such cells by voltage, which makes the need for in-system balance
management more critical for this chemistry. But such a system clearly cannot
be based just on voltage in this case. Later we will look at systems that work for
flat chemistries.
The second important reason for SOC differences between the cells in a
pack are differences in self-discharge rates between the cells. The self-discharge
rate is strongly dependent on temperature. It approximately doubles with every 10C increase from room temperature. System design does not always take
into consideration the need to heat the pack evenly, and places various heatgenerating components such as the application processor, backlight, and memory in such a way that they can fit most of the components into the smallest

114

Battery Power Management for Portable Devices

Figure 4.1 (a) OCV dependence on SOC. (b) OCV differences at different states of charge
between two cells with a SOC imbalance of 1%.

space, rather than ensuring that cells in a pack will have the same temperatures.
This will cause one cell that is hotter to leak more charge than a cell that is
cooler. This cells SOC will gradually decrease, because the charger puts the
same amount of coulombs into each serially connected cell, but some coulombs
in the hot cell get internally short-circuited due to self-discharge and do not
contribute to increasing SOC. This process is quite slow, because even at the
highest temperature of 60C, the self-discharge of a Li-ion battery is only about

Cell-Balancing Techniques: Theory and Implementation

115

50% in a year. But over time, the difference in SOC can become significant
because the effect is accumulative (the same cell will usually be hotter than the
others). Battery pack design could include a heat spreader between the system
and the pack, and a good cooling surface on the other side of the cells to keep
their temperature and self-discharge in general to the minimum. That also helps
to decrease cell degradation, which also happens to be strongly accelerated by
temperature.
A SOC imbalance can be also caused by uneven leakage to the battery
pack circuit from different cells. It is easier to power pack electronics from the
low voltage of just one cell (usually the one closest to the pack ground), because
cheaper low-voltage ICs can be used. It looks like a nice shortcut to take, because the electronics would only consume some hundreds of microamps, which
is negligent compared to multi-ampere-hour size cells. Unfortunately the effect
is accumulative. Over a long period of time, the difference in the amount of
charge removed from the lowest cell will increase and eventually reach substantial numbers. For example, if we have three 2,200-mAh cells (Qmax), and
discharge one by 100 mAh (Q1), the second by 100 mAh, and the third by
100mAh of actual load + 100 mAh accumulated from 1 month of low current
due to powering some circuit, the first and second cells chemical state of charge
will be (Qmax Q1)/Qmax = 95.4%, but the third cell will be at 91%. So we can
say that cell 3 is imbalanced by 4.4%. This in turn will result in a different open
circuit voltage for cell 3 compared to cells 1 and 2, because the OCV is in direct
correlation with the chemical state of charge. This problem can be resolved by
good pack design, which powers pack electronics only from the entire pack
voltage. It also assures that all of the connections to each serial cell (for example,
those used for voltage measuremetns) are high-ohmic. Typically, these connections are ADC inputs with impedance in mega-ohm ranges, and do not draw
more than a microamp or two. In addition, this impedance is the same for all
cells, so even the tiniest current drawn does not cause an imbalance. All major
safety and gauging ICs from Texas Instruments, Intersil, or Maxim are powered
from the pack voltage and avoid this issue.
4.2.2 Total Capacity Differences

Sometimes cells have the same voltage at the end of a charge and, hence, appear to be perfectly balanced (and correspondingly have the same SOC), only
to show a very large deviation of voltage at the end of discharge. Inversely, cells
having an equal SOC and voltage in the discharged state can show large differences at the end of charge. What is the reason for this mystery? It happens to be
caused by differences in cell capacities. Indeed, if you have two cups of different
heights, they will both be equally empty (e.g., state of charge zero and perfectly
balanced) when there is no water in them. But once you pour an equal amount

116

Battery Power Management for Portable Devices

of water into both cups, the taller will be half full, but the shorter will be completely full (e.g., states of charge 50% and 100%, respectively).
The dilemma in such a situation is that you cannot just balance cells with
different capacities once and have them stay balanced during charge/discharge
(as was the case with SOC balancing for the cells of the same capacity when not
given enough time for self-discharge). We have four choices:
1. Balance the cells on top (in the fully charged state) and let them diverge at the bottom. This will be repeatablethey will always stay
balanced on the top after that without any further action.
2. Balance the cells on the bottom (in the discharged state) and let them
diverge at the top. Again, once the initial balancing is done, it will be
a persistent state.
3. Completely rebalance the cells during each discharge, for example,
extract energy from higher capacity cells on the way down so that
by the time it reaches the bottom all cells have the same energy. This
would also require a complete rebalancing during each charge, so that
the moment the energy from smaller capacity cell becomes full is delayed until other, larger capacity cells become full. Going back to our
analogy with the cups, this would be equivalent to leaking some water
from the jar past the shorter cup when poring it so that it will be at the
same level as the taller cup at the end.
4. Finally, there is a variation of choice 3, where we are not just throwing
away the energy that we take from the larger cell on the way down, but
are actually passing it to the lower capacity cells. Inversely, on the way
up, energy from smaller capacity cells is not just dissipated, but forced
into larger capacity cells. This way, in our example with the cups, we
will not end up with two cups at just the 50% level (e.g., matching
larger cup), but with both cups at some intermediate higher level, say
75%, because we did not spill any water on the table. It also helps with
reducing the subsequent cleaning.
The best choice appears to be obvious (choice 4), but in reality it is not so
simple. Because nothing in life is free, passing energy from one cell to another
requires an actual hardware implementation (discussed below). Lets just say
that it has inductors (or capacitors) and high-power FETs so it is about as large
as all of the safety circuits and gauging circuits taken together, so you would
likely double the size and cost of your overall battery management solution.
Before making a heavy investment in additional hardware, we have to consider
how much benefit will we get. That will depend on several factors:

Cell-Balancing Techniques: Theory and Implementation

117

1. How much cell capacity imbalance is there? Little imbalance, little

benefit.
2. How large is the cell? Sometimes it is so large and expensive that the
additional cost of the balancer would be justifiable to get all of the
energy from such a large system.
3. Is heat dissipation (e.g., spilled water on the table) a problem in this
particular design because batteries are large and space is confined?
Actually, for many laptop designs it turns out that capacity differences
after factory grading can be reduced to 1% to 2%. So in many cases it is cheaper
to use a 1% to 2% larger battery rather than investing in doubling the size and
cost of hardware.
Now, if we are not going to actually transfer capacity from cell to cell,
what is the next best option? Well, closer observation shows that option 3 requires a lot of spilling of water both on the way up and on the way down,
and it has to happen for every single cycle. Heat dissipation due to damping of
additional energy into the environment can be undesirable because it requires
installation of additional cooling capacity. But most importantly, what do we
gain if we completely rebalance the cells on the way down? All cells will reach
0% at the same time but this will be the same time as the lowest capacity cell
would reach it without any balancing. A battery pack protector would have shut
down the discharge of the pack at the same time anyway, because the lowest
capacity cell will trigger an undervoltage condition. So we released all of this
heat from higher capacity cells (at the cost of some additional hardware!) and
improved run-time by zero seconds.
How about the charge direction? There is some benefit to keeping cells
from exceeding the recommended charging voltage, so if we can prevent the
lower capacity cells from going all the way to the overvoltage threshold (which
can be as high as 4.3V for 4.2V normal charging voltage cells), we could improve their health and stop the vicious cycle where the pack gets overcharged,
degrades more, ends up with even lower capacity, overcharges more, degrades
more , repeat from the beginning. But do we have to do it every cycle? Actually, if we give up on the idea of rebalancing the cells on the way down (which,
as we have seen, has no merit) then we do not have to rebalance them on the
way up either! We can just balance it once, on the top, and it will stay this way
without any further action! Which brings us to choice 1, which is indeed the
most popular choice in portable electronics. Choice 1 is often called top balancing and choice 2 is often called bottom balancing. To summarize, bottom balancing has no merit in systems that utilize energy dissipative balancing, whereas
top balancing helps preventing excessive degradation of a lower capacity cell.
But wait, do you have this question on your mind: If we need to balance
only once, why do we even need the balancing circuit in the pack? We could

118

Battery Power Management for Portable Devices

just balance cells on the top to correct the capacity differences during assembly
(e.g. gross balancing) and be done with it, all the while saving money and
space. Well, this statement indeed makes sense. We really do not need an expensive and spacious high-current balancer designed to provide such high currents
that would allow keeping up with high rate external loads and compensating
cell capacity differences in real time. We still need a small lower power balancer
to continuously counteract other causes of SOC imbalance that are themselves
continuous and accumulative in their naturethe differences in self-discharge
rate of cells due to their different temperature, for example.
In addition, the cells degrade at a slightly different rate, so cells that had
the same capacity initially will gradually develop some individuality. So this
additional discrepancy has to be balanced out, as it develops, in the pack itself.
However, both the self-discharge rate and degradation rate are extremely slow
and for that reason only balancing current in the milliamp range is needed to
keep it in check. Such low current needed for maintenance balancing can be easily provided by integrated circuits that are already there, such as protector circuits, without any added cost or size. We look at some specific examples shortly.
4.2.3 Impedance Differences

In addition to SOC differences, another cause of voltage discrepancies among

cells is the cell impedance difference. However, this has an effect only if current
is flowing. How large is the effect? It depends on cell impedance differences and
the current. Internal impedance differences among the cells can be expected in
the ~15% range in the same production batch as can be seen form the example
test data shown in Figure 4.2(a).
Impedance imbalances do not cause differences in fully relaxed OCV,
when any effect from current flow is already dissipated. However, they will

Figure 4.2 Impedance spectra differences between 50 cells in one batch for manufacturer
(a) and 50 cells for manufacturer (b). Data shown range from 1 kHz (left) to 10 mHz (right).

Cell-Balancing Techniques: Theory and Implementation

119

cause differences in cell voltage during discharge. For steady-state current flow,
cell voltage can be approximated as V = OCV + I * R, where R is the lowfrequency portion of cell internal impedance (right side of the graph in Figure
4.2). If the current is negative (discharge), the voltage will be lower for a cell
with higher R. If the current is positive (charge), the voltage will be higher for
a cell with higher R.
As can be seen from Figure 4.2, high-frequency impedance at 1 kHz (left
side of the graph) is well matched for manufacturer (b) and not as well matched
for manufacturer (a). However, low-frequency impedance (the one that will actually matter for continuous discharge) is equally badly matched for both. The
reason for this is that most cell makers have access to simple 1-kHz impedance
meters that allow them to grade the cell based on high-frequency impedance.
High-frequency (1-kHz) measurement is very fast and allows for the detection
of massive failures such as a short circuit or current collector disconnect. However, it is not very useful to observe the whole range of electrochemical properties of the battery related to actual charge storage. Low-frequency impedance
would be more useful for preventing a voltage imbalance, but its measurement
takes at least 10 seconds, so it is very rarely used in cell production (but could
be used by a pack maker for improving the cell matching).
No balancing algorithm can eliminate the resistance differences; they are a
permanent property of a battery pack once assembled (that is why preassembly
grading is beneficial). In fact, the imbalance can increase with aging because
different cells impedance is likely to change at a slightly different rate. But impedance differences need to be considered in any balancing scheme, especially
one based on voltage, because they can significantly distort attempts to balance
what we can and should balance; namely, the SOC. Note in Figure 4.3 that for
the absolute majority of discharges (from 10% to 100% SOC) the distortion
of voltage that is caused by the impedance deviation is larger than that caused
by a SOC imbalance.
By looking at voltage alone we cannot distinguish which part of the cell
deviation is due to a SOC difference and which part is due to an impedance
difference. Both parts can shift voltage in the opposite direction! If we do not
know about this and just assume that all voltage shift is due to a SOC difference, we might be tempted to correct it by bypassing some charge through the
cell with the higher voltage that appears to have a higher SOC.
However, if most of the difference is caused by an impedance imbalance
(as is commonly the case), bypassing more current through this cell will result
in the opposite effectit will increase the SOC difference from other cells to
a larger value than it would be without balancing. As a result, the open circuit
voltage of this cell at the end of charge will be different from the other cells and
can reach high levels, potentially causing the safety circuit to trip.

120

Battery Power Management for Portable Devices

Figure 4.3 Solid line: voltage differences between two cells with 15% impedance imbalance
at C/2 discharge rates. Dotted line: difference between the cells with 1% SOC imbalance for
comparison.

If, for example, we are using a simple balancing scheme when a bypass
FET allows us to turn on a load parallel to any cell and the bypass FET is
turned on based on voltage during charge, it can cause an actual increase of the
imbalance through bypassing the cell with the higher impedance. At the end
of the charge, the IR rise becomes insignificant because of current decrease, so
that the FET switches on at the other cell. However, it happens too late so at
the end of charge this procedure results in a higher SOC and higher voltage for
low-impedance cells. Eventually it will lead to increased cell degradation. This
problem can be reduced if cell balancing only switches on near the end of the
charge when the current is reduced and so the I*R drop has a smaller effect on
battery voltage.
It is more difficult to fight the IR effect during discharge, because there
are usually no predictable periods of low current except for some rare application. Fortunately top balancing (which we earlier found to be the only one
useful for dissipative implementation) is mostly done only during charging.
There are periods of low current (e.g., in the case of inactivity, rest) that
potentially could be used for balancing without IR effect, but it would be still
beneficial only if balancing could be done that would ensure the cells were equal

Cell-Balancing Techniques: Theory and Implementation

121

on top (at the end of charge), which requires predictive balancing rather than
reactive balancing unless the rest is happening in the fully charged state. See
the later discussion on balancing algorithms for more details.
Note that distorting due to impedance differences is inherent to any voltage-controlled balancing method (such as for inductive energy redistribution)
and not just to simple bypass balancing. In fact, in energy redistribution cases,
the effect can be even more dramatic because energy will be passed back and
forth with the high currents that such balancers are typically capable of, which
can cause overheating, loss of efficiency, and even complete instability of the
control circuit.
Another effect of battery impedance on voltage imbalance exists regardless of any impedance differences. This effect just amplifies voltage differences
due to a SOC imbalance. Again, to explain the effect we can model the voltage
under steady-state load as V = OCV(SOC) + I * R(SOC) (considering that
discharge current is negative). Because function R(SOC) is rapidly increasing
its value as SOC approaches zero, the voltage differences between the cells with
fixed SOC imbalance increase in highly discharge states, as shown in Figure
4.4. This gives the impression that there is an increased need for balancing near
the end of discharge. However, if the SOC imbalance is removed during other
stages of discharge and is absent by the time low SOC is reached, the increased

Figure 4.4 Voltage differences under C/2 load at different states of charge among cells with
a 1% SOC imbalance. Solid line: Differences for OCV case for comparison.

122

Battery Power Management for Portable Devices

voltage differences near the end of discharge will be eliminated without need
for high bypass currents.

4.3

Effect of Imbalancing on Performance

4.3.1 Premature Cell Degradation Through Exposure to Overvoltage

Now that we have a clear understanding of the underlying mechanisms of voltage differences, we can evaluate in more detail different issues that some of these
mechanisms can cause. Impedance differences will cause cell voltage differences
during charge and discharge, but because this difference in voltage is purely
ohmic in origin (to large extent), just like IR drop/rise across a resistor, it does
not cause accelerated cell degradation as such and will disappear once the current is turned off. So although there is nothing we can do about this impedance
difference, there is also no need to do anything. On the other hand, in the cases
of a SOC or total capacity imbalance, the cell with the higher resulting SOC is
exposed to higher chemical potentials that will cause accelerated degradation.
For example, what happens if one cell has less capacity than the other three
serially connected cells in the pack, if they all start at the same state of charge?
CC/CV charging will bring the pack to 4.2 4 = 16.8V (typical). However,
individual cell voltages will not be equal. As you can see in Figure 4.5, the lowcapacity cell will have a much higher voltage than the remaining cells, while
the normal capacity cells will have a lower voltage than is achieved in normal
charging. As the cell is exposed to higher potential, it will degrade more, thus

Figure 4.5 Individual cell voltage versus capacity deficiency from nominal.

Cell-Balancing Techniques: Theory and Implementation

123

increasing the capacity deficiency, which will move the pack to the right on
the graph in Figure 4.5. Eventually, when the lower cell reaches a total capacity
deficiency above 10%, its cell voltage rises into the dangerous area above 4.3V,
which will result in extreme degradation of this cell or even become a safety
concern.
Note that not all battery chemistries are equally affected by cell voltage
imbalances at the end of charge. While the Li-ion chemistry is especially vulnerable because of its ability to store almost 100% of all energy delivered with
negligible self-discharge in its operational voltage range, lead-acid, NiMH and
NiCd chemistries are relatively tolerant to overcharge because they can respond
to increased voltage by internal shuttle reactions that are equivalent to a chemical short circuit inside the cell. For example, in a NiMH battery oxygen and
hydrogen generated after the end of charge recombine inside the cell, building
water. This causes extensive heating because all the energy of the charger is
converted to heat rather than stored, which is undesirable but at least it does
not cause thermal runaway. Still, overcharge at high rates does cause increased
pressure inside the cell and will accelerate cell degradation and can even create
a chance for explosion or venting. The need for cell balancing has to be evaluated in conjunction with rate capability, cooling, and other properties of the
charging system.
4.3.2 Safety Hazards Resulting from Overcharged Cells

Li-ion batteries have very high electric energy concentrated in a small volume.
While the possibility of its release via a short circuit can be prevented by appropriate mechanical protections, the coexistence of highly reactive chemicals in
proximity makes this battery inherently dangerous. Overcharging and overheating of the battery cause the active components to react with electrolyte and with
each other, ultimately causing an explosion and fire. Thermal runaway can be
caused merely by overcharging a single cell to voltages above 4.35V. Other cells
of the pack will also join the explosive chain reaction if one cell is compromised.
That is why continuous cell balancing should prevent any cells from getting
anywhere near the dangerous voltage territory, and a safety protection circuit
should terminate the charge if this somehow happens.
4.3.3 Early Charge Termination Resulting in Reduced Capacity

Additional safeguards present in the battery pack primary protectors (such as

bq30z55) and the independent secondary protectors can help alleviate safety
issue. The protector and gas-gauging IC will terminate charging if one of the
cells voltages exceeds the programmable cell overvoltage threshold (default
4.35V). An overvoltage protector will terminate charging and prevent an unsafe
condition from occurring, but at the same time it will keep the whole pack un-

124

Battery Power Management for Portable Devices

dercharged (since other cells are now in a much lower state of charge). Overall,
the pack loses energy as the result because the higher voltage in the steeply rising
area close to a fully charged state does not compensate for the large capacity loss
in much more flat areas where the mAh/volt are much higher.
Because the effects of cell degradation caused by imbalance are autoaccelerating, preventing such a vicious cycle allows us to extend battery pack
life significantly and to provide a longer run time despite some initial capacity
imbalance.
4.3.4 Early Discharge Termination

So far we have mostly discussed cell degradation if overcharged. This is the

greatest safety concern and also the most common degradation mechanism.
However, the cell will also be damaged if it is severely overdischarged. In fact,
if cell voltage goes below 2V, the actual dissolution of the Cu-current collector
will occur. This is deadly to the cell if the process is allowed to continue to a
significant extent (no current collector, no charge/discharge). For this reason
any device using a Li-ion battery absolutely has to have a discharge termination
based on voltage. Some nave attempts to introduce a consumer-replaceable
Li-ion battery, for example as a drop-in replacement for coin cells, have so far
faltered for the very reason that primary cells that are being replaced did not
require any undervoltage shutdown (they just die, being primary) and might
keep draining substantial current regardless of battery voltage even when no
useful operation is taking place. If a Li-ion cell is placed into a device that was
not designed for it, the cell will be overdischarged below 2V and die just like the
primary cell that it replaces.
This is mostly not an issue for devices designed to use Li-ion cells, because
they will have a full-off state when the pack voltage goes below a safe threshold.
In addition, the battery pack itself will often have a battery management system.
To prevent overdischarge of cells and resulting damage, battery managements
system will terminate the discharge process if any of the cells reach the predetermined low-voltage threshold. The cell-based termination voltage is usually
set to a lower value than the pack-based threshold divided by number of series
cells, so that the difference can allow for a small imbalance. For Li-ion batteries,
the threshold varies from 2.7 to 2.2V depending on the typical discharge rate.
However, a larger imbalance will cause the overall pack to terminate when a cell
with lowest capacity or SOC will reach cell undervoltage while other cells still
have energy left.
Redistributing the energy from higher cells to lower cells during the end
of discharge phase can increase a batterys useful discharge time. Note that if
bottom balancing is being used, it needs to be a redistribution of energy, and
not just a bleeding out of extra charge; otherwise, it would just dissipate heat

Cell-Balancing Techniques: Theory and Implementation

125

and not increase the run time. To be effective, an inductive or capacitive energy
redistribution circuit with high efficiency (usually above 80%) is required. If
the control algorithm reacts only to voltage differences that become noticeable
only very close to the end of discharge, it would also require a high-rate bypass
capability to keep up with the high discharge current. Such circuits are expensive to implement in redistribution balancing circuits, and larger inductors and
FETs use up space that is at premium in portable devices. An approach that
utilizes the hardware more effectively would be to gradually redistribute any existing SOC imbalance during the entire charge/discharge process, not just when
it results in acute voltage differences (at the end of discharge); this is known as
predictive balancing. This, of course, requires the ability to determine how much
charge needs to be bypassed somehow without relying only on voltage. How to
accomplish this is discussed in the balancing algorithms section.

4.4

Hardware Implementation of Balancing

4.4.1 Current Bypass

One simple implementation of cell balancing uses a MOSFET in parallel with

each cell and controlled by a comparator output for simple voltage-based algorithms that turn on the bypass FETs during the onset of voltage differences, or
controlled by a microcontroller for more complex and effective algorithms that
can work continuously regardless of variations in the voltage. A general setup is
shown in Figure 4.6.
The main choice here is to use MOSFETs that are integrated in the balancing controller IC and typically have bypass currents from 9 to 2mA (depending

Figure 4.6 Cell-balancing setup using bypass FETs.

126

Battery Power Management for Portable Devices

on the choice of the external resistors), or to use external FETs with bypass
capability that can be freely tailored to particular application needs.
In Li-ion batteries that have a very low self-discharge capability and, therefore, an accumulative imbalance per cycle of usually less than 0.1%, the bypass
current of internal FETs is sufficient to keep the pack continuously balanced.
In other chemistries where self-discharge rates are much higher and, therefore,
differences in the self-discharge rates among the cells result in higher SOC differences per cycle, higher rates might be needed. Some balancing circuits have
separate pins for voltage measurement and charge bypass; however, this is not
common in portable devices because the larger number of pins increases the
size of the device, which is a disadvantage in constrained spaces. The issue with
balancing current interfering with voltage measurement is usually addressed
by the firmware turning balancing FETs on at a time when measurements are
already finished.
Passive cell balancing using integrated FETs is limited by low balancing
current and, therefore, may require multiple cycles to correct a typical imbalance. To achieve fast passive cell balancing, an external bypass circuit can be
implemented by modifying the existing hardware. Figure 4.7 shows a typical
implementation. The internal balancing P-MOSFET SN for a particular cell,
which needs to be balanced, is turned on first. This creates a low-level bias current through the external resistor dividers, R1 and R2, which connect the cell
terminals to the battery cell balance controller IC. The gate-to-source voltage
is thus established across R2, and the external MOSFET SEN is turned on. The
on-resistance of the external MOSFET SEN is negligible compared with the
external cell balance resistance RBAL, and the external balancing current, IBAL, is
given by IBAL = VCELL/RBAL .
By properly selecting the RBAL resistance value, we can get the desirable
cell-balancing current, which could be much higher than the internal cell-balancing current and can speed up the cell-balancing process.
The drawback of this method is that balancing cannot be achieved on adjacent cells at the same time, as shown in Figure 4.8. When internal MOSFETs
SN and SN+1 of the adjacent cells are turned on, there is no net current flowing

Figure 4.7

External passive cell-balancing circuit.

Cell-Balancing Techniques: Theory and Implementation

Figure 4.8

127

The adjacent cell-balancing issue.

through R2 and no voltage drop is created across R2. So there is no gate-source

bias voltage of MOSFET SEN, which remains off even when the internal MOSFET SN is on. In practice, this is not an issue because the fast external cell
balancing can quickly balance the cell associated with SEN2, and then the cell
associated with SEN1 will be balanced. Therefore, the adjacent cells cannot be
balanced at the same time, and only every other series cell can be balanced with
this approach at the same time. However, some cell balancing ICs have two
pins per cellone for voltage sensing and one for balancingwhich makes for
a simpler circuit that can balance all cells simultaneously, but such ICs handle
fewer cells or have a larger number of pins.
4.4.2 Charge Redistribution

The disadvantage of the current bypass approach is that the energy of the bypassed charge is wasted. While this can be acceptable during charge while the
system is connected to a power grid, during actual usage of the battery in portable applications every milliwatt-hour is precious. This makes desirable a cellbalancing approach that would allow us to drain the high cells to the bottom
in the most efficient way.
The ultimate approach to accomplish this is to use a pack that has no serially connected cells at all. The step-up converter then ensures that the device
obtains sufficient voltage. This way, energy waste as a result of the cell-balancing process is completely eliminated. The trade-off, however, is lower efficiency
of the power supply, as well as increased size and complexity.
Other solutions can include circuits that allow to for the transfer of energy from high cells to low cells rather than burning it in a bypass resistor. Note
that use of the correct control scheme is still critical even in this case because

128

Battery Power Management for Portable Devices

all circuits have limited efficiency and if a charge is unnecessarily shuttled back
and forth multiple times due to, for example, an IR effect on voltage, overall
efficiency could go close to zero after multiple swings even if the single-pass
redistribution efficiency can be as high as 80%. So all of the balancing algorithms discussion in subsequent chapters apply to charge redistribution circuits
just as they apply to bleed balancing.
4.4.3 Charge Shuttles

One simple approach for redistributing the energy among cells is to connect a
capacitor first to the higher voltage cell, then to the lower voltage cell, as shown
in Figure 4.9(a). More complicated implementations allow us to connect not
only two nearby cells, but also several series cells, as shown in Figure 4.9(b).

Figure 4.9 (a) Simple capacitor-based shuttle cell balancing circuit. (b) Charge shuttle circuit with several series cells.

Cell-Balancing Techniques: Theory and Implementation

129

Cell 1, cell 2, , cell n share flying capacitors with their two neighboring cells,
so charge can travel from one end of the cell string to the other. This approach
would take a large amount of time to transfer charge from the high cells to the
low cells if they are on the opposite ends of the pack because the charge would
have to travel through every cell with time and efficiency penalties. This would
not be an efficient solution.
Energy loss during capacitor charging is 50%, so heating in the FETs used
as switches has to be considered if high-current balancing is supported. However, because there is no charge loss with this process, the energy available on
the pack terminals decreases only due to the decrease of cell voltages. Another
problem is that high voltage differences between the imbalanced cells exist only
in highly discharged state. Because this method transfer rate is proportional to
voltage differences, it only becomes efficient near the end of discharge or the
end of charge so the total amount of imbalance, that can be removed during
one cycle, is low.
4.4.4 Inductive ConverterBased Cell Balancing

Active cell balancing overcomes the energy loss of the passive approach by using
capacitive or inductive charge storage and shuttling energy to deliver it where it
is needed most. This can be done with minimum energy loss if combined with
an optimal balancing control algorithm that allows us to take full advantage of
a circuits inherent redistribution efficiency by avoiding back-and-forth shuttling. It is preferable for efficiency-conscious designs and for applications where
delivering the maximum run time is top priority.
A switch-mode power converter concept can apply to the cell balance for
achieving the energy transfer from one energy source to another. Figure 4.10
shows the active cell-balancing circuit based on the switch-mode power conversion concept.
A MOSFET, a diode, and a power inductor are composed of a buckboost converter to complete a charge transfer between an adjacent pair of cells
as shown in Figure 4.10. This is a bidirectional buck-boost converter, which
can transfer cell energy from either direction. Figure 4.11 shows the switching waveforms of the inductor current and cell-balancing current. If the cellbalancing control algorithm determines that the top cell N needs to transfer its
energy to the lower cell, the SN signal, operating at a few hundred kilohertz
with a certain amount of duty cycle triggers to turn on the P-MOSFET SN. The
voltage of top cell VCELLN applies to the inductor LN and the inductor current
linearly increases. The cell energy is first transferred from the top cell to the
inductor during this time period. When the SN signal resets, SN is turned off
at t1, and the energy stored in the inductor reaches a maximum value. Because
the inductor current must flow continuously, the diode DN1 is forward biased

130

Battery Power Management for Portable Devices

Figure 4.10 Buck-boost converter based cell-balancing circuit implemented in a bq76PL537A active cell-balancing IC.

Figure 4.11 Switching waveform of active cell-balancing circuit.

and a negative cell voltage VCELLN1 is applied to the inductor, which results in
an inductor current decrease and transfer of the energy stored in the inductor to
the lower cell. When the inductor current reaches to zero at t2, all energy stored
in the inductor has been completely transferred to the lower cell, and the diode
is naturally turned off with the minimum loss. If the cell N 1 has more energy

Cell-Balancing Techniques: Theory and Implementation

131

than that of the top one and needs to transfer its energy to cell N, switch SN1
is turned on first and the energy from cell N 1 is stored in inductor L. When
switch SN1 is turned off, then the energy stored in the inductor is transferred to
the top cell N through the diode DN. In this energy transfer process, the energy
loss includes loss from the series resistance of the inductor, and the diode, and
switching loss of the MOSFET.
Overall, 90% power transfer efficiency can be achieved with such active
cell balancing. The balancing current is determined by the inductance switching period and its turn-on duty cycle. The current level could be much higher
than passive cell balancing and more efficient. Besides the obvious advantages,
the beauty of such cell-balancing technology is that balancing is achievable regardless of the individual cell voltages.
Figure 4.12 shows the active cell-balancing circuit with N series cells.
From this circuit, it is found that the energy can only be transferred from the
top cell to the lower adjacent cell or from the lower cell to higher cell as well.
As we know the flyback converter is the isolated power converter of the
buck-boost converter. Its output is isolated from the input, and output can be
floating such that it can connect anyplace. Figure 4.13 shows an active cellbalancing circuit that can transfer the energy from the bottom cell to the top
cell directly with a flyback converter. When switcher S1 is turned on, the bottom cell voltage is applied to the primary winding. The current flowing through
the magnetizing inductor linearly increases and its energy is stored in the magnetic field. When switcher S1 is turned off, the energy stored in the magnetizing inductor is released to charge the top cell through the output diode Dn .
Therefore, the extra energy from the bottom cell can be transferred to the top
cell through a flyback converter. The main limitation of such an active cellbalancing method is that energy is only distributed to the adjacent cell, not to
any target cell. On the other hand, such cell-to-cell balancing is only good for
battery packs with few cells in series. For long strings, due to the inefficiency
of the converters at each step of the transfer, too little energy can be transferred
from one end of the pack to the other, making active balancing less efficient
than dissipative (passive) balancing.
How do we charge the weak cell or discharge the strong cell to achieve
cell balancing? After measuring the voltage or capacity for all cells, the average voltage or capacity can be calculated. The cell with the biggest deviation
from the average could be identified. The cell with the lowest voltage or capacity can be recharged from the pack, while the cell with the highest voltage
or capacity could be discharged. Figure 4.14 shows the synchronous switching bidirectional flyback converter used to achieve active cell balancing for any
target cell. Assume cell 2 is recognized as the weakest cell, and it needs to be
charged. Switcher SP is turned on first, the whole pack voltage is applied to the
primary winding, and it stores the energy in the magnetizing inductor of the

132

Battery Power Management for Portable Devices

Figure 4.12 N-series cell-balancing circuit.

transformer. When switcher Sp is turned off, the MOSFET S2 associated with

the selected weakest cell is turned on. The stored energy of the transformer can
be transferred to cell 2. This can extend the battery packs run time by preventing the weakest cell from reaching the end of discharge earlier than that of the
rest of cells. Therefore, this operation can achieve the cell balancing required for
transferring energy from the pack to any either of the cells.
On the other hand, if is more effective to shift energy from a strong cell
to the pack. Without cell balancing, the charge process has to stop immediately
when one cell reaches it maximum cell voltage, even though the other cells
have not been fully charged. Assume that cell N 1 has been detected as the
strongest cell in the battery pack, and we need to discharge the energy from cell

Cell-Balancing Techniques: Theory and Implementation

133

Figure 4.13 Flyback-based cell-balancing circuit for transferring the energy from the bottom
cell to the top cell.

N 1 and redistribute it into the pack. MOSFET SN1 is turned on first. The
cell voltage is applied to the winding connected to cell N 1 and stores the
energy from the cell N 1 in the magnetic field. Once MOSFET SN1 turns
off, MOSFET Sp, connected in the primary winding, is turned on so that the
energy stored in the magnetic field is transferred back into the pack through the
primary winding. So, this operation can achieve cell balancing by transferring
cell energy to the pack.

4.5

Balancing Algorithms

Regardless of the particular hardware implementation, there is always a decision

to be made regarding when to turn on a bypass switch or when to engage the
energy exchange circuit for a particular cell. Different algorithms used to make
this decision are reviewed in this section. For simplicity we will refer to the case
of current bypass because application of the logic to other balancing schemes
is trivial.

134

Battery Power Management for Portable Devices

Figure 4.14 Synchronous switching bidirectional flyback converter-based active cell-balancing circuit.

4.5.1 Cell Voltage Based

The simplest algorithm is based on the voltage difference among the cells. If
that difference exceeds a predefined threshold, bypass is engaged. To resolve several problems that accompany this simple method, more complicated modifications can be implemented if a microcontroller is used to execute the algorithm:
Balancing during charge only is used to save energy in portable applications.
Balancing at high states of charge only is used to decrease the effect on
SOC balancing that can come from an impedance imbalance, because
current decreases during the CV mode until charging terminates on
minimal taper current.
A variation of above, in which the balancing is enabled only at a low
current regardless of the SOC.

Cell-Balancing Techniques: Theory and Implementation

135

Simultaneous multicell balancing makes decisions about which cells have

to be bypassed in terms of the entire pack, not just neighboring cells, as
is the case with comparator-based solutions.
One advanced implementation of voltage-based algorithms that incorporates all of the above optimizations is used in the bq2084 battery fuel gauge.
Figure 4.15 shows the voltage convergence of multiple cells during balancing.
4.5.2 SOC Based

If a method for determining SOC that is independent of the voltage being under load is available, the balancing algorithm can be improved, because it is no
longer vulnerable to impedance variations. However, if no independent method
to measure each cells full capacity exists, equal capacity has to be assumed for
all cells such that the capacity imbalance will not be considered. Such a method
works as follows:
1. Determine the initial SOC for each series cell bank separately. One of
the determination methods is to use an open circuit voltage correlation with the state of charge. This method can only be implemented
in a microcontroller with flash memory and significant computational
resources because of the need to evaluate voltage versus SOC function
OCV(SOC,T) in real time.
2. Determine how much charge is needed for each cell to reach a fully
charged state. This requires knowledge of total capacity, which is assumed equal for all cells.
3. Find the cell that has the largest amount of charge needed to reach full
capacity, and find the differences dQ among all other cells that need a
charge and that of the largest one.
4. This difference has to be bypassed for each excessive cell during one
or multiple cycles. To achieve that, the bypass FETs are turned on during charging for the duration of each cells calculated bypass time. The
bypass time is calculated dependent on the value of the bypass current,
which in turn depends on values of bypass resistance, R_bypass, because time = dQ * R_bypass/(V_average * duty_cycle). Although not
a very accurate estimate of needed bypass time, this method is acceptable for low-rate bypass, because during many cycles of balancing, the
amount of needed balancing time will be recalculated after every cycle.
5. Alternatively, the bypass current around each cell can be continuously
integrated I = Vcell/R_bypass) and bypass FETs are turned off once

Figure 4.15 Cell open circuit voltages of a four-cell pack at the end of charge during balancing. Initial imbalance = 10%.

136
Battery Power Management for Portable Devices

Cell-Balancing Techniques: Theory and Implementation

137

the needed bypass charge has passed. This method is preferable if high
bypass currents, capable of balancing cells in one cycle, are used.
4.5.3 SOC and Total Capacity Based

Further improvement of the above method takes into account differences in

total cell capacities (Qmax). This becomes possible if the cell-balancing algorithm is an integral part of the more complex gas-gauging algorithm that is
monitoring the state of each cell and capable of measuring changes in the total
cell capacity of each cell, as is the case, for example, with bq20z80 cell balancing. Overall, the balancing method is similar to that described in Section 4.5.2,
except for calculation of dQ, which now takes into account each cells individual
capacity. Figure 4.16 shows the progress of the cell-balancing process, which
causes changes in the open circuit voltage at the end of charge in a three-cell
pack, where cell 1 was individually discharged and cell 2 was charged by 2%
prior to the test. The bypass resistance used in this test was 700 .

4.6

Summary

In this chapter, we analyzed three main mechanisms that can cause voltage
differences among cells that are serially connected in a pack: (1) a SOC imbalance, (2) a total capacity imbalance, and (3) an impedance imbalance. Because
the ability to add or remove only a certain amount of charge is available to

Figure 4.16 Evolution of cell voltages during SOC/Qmax balancing, starting from an initial 2%
down (cell 1) and 2% up (cell 2) imbalance.

138

Battery Power Management for Portable Devices

balancing algorithms, only the first type of imbalance, an SOC imbalance, can
be eliminated. The second type of imbalance (capacity) has to be taken into account in bypass charge calculations and the third (impedance) should be kept in
mind as a distortion, if voltage is used as the balancing criteria, to improve the
balancing process and prevent the introduction of a larger imbalance.
Considering the low self-discharge rate of Li-ion cells, we can conclude
that if continuous balancing is engaged, the use of integrated FETs provides
sufficient balancing current. Use of external FETs may be required if voltagebased balancing is used; that is only active in the areas where a SOC imbalance
is reflected by high voltage differences (mostly at the end of discharge). Active
balancing methods can provide higher efficiency, but are not at present cost
effective for portable applications. An exception might be the case in which
extremely high reliability and longevity of the battery pack are needed, because
active balancing extends the usable life of a pack, primarily due to the complete
use of pack energy regardless of the amount of imbalance.
Voltage-based balancing algorithms have the advantage of simplicity of
hardware and implementation, but suffer from slower balancing rates and the
possible introduction of additional imbalance through distortions from impedance differences. SOC and total capacity based methods are more complicated
to realize but can take advantage of the already present gas-gauging capability
of controller ICs and ensure, for given bypass capabilities of the hardware, the
most accurate and fastest balancing possible.