Lec 18: Transistor sizing for combinational circuit - II

IIT Roorkee July 201
61 followers
Aug 7, 2024
This lecture covers the CMOS inverter logical effort for multi-V_dd operations.
[Music]
[Music]
welcome to this lecture so this is in
the module minimum energy point for low
voltage operation and transistor sizing
for near threshold circuits and in this
lecture we will be discussing transistor
sizing for combinational circuits and
simple back of the envelope calculation
method for estimating logical efforts of
near threshold voltage circuits so we
will first discuss the charging and
discharging mechanism of uh near
threshold voltage circuits particularly
logic gates which have stacked
transistor
and then we will discuss a very simple
method to obtain an approximate value of
logical effort of these near threshold
logic gates and then we will discuss the
logic path sizing transistor sizing in
logic path using this modified logical
effort method so first we will discuss
how do we get how do we estimate WP by W
now what is meant by WP by WN suppose we
want to balance the the transition times
of an inverter so you balance them by
adjusting the ratio of the widths of n
moset and p moset lengths are kept at
the minimum Channel length so why do you
do this because you may want sometimes
want equal rise and fall transition
times sometimes you may want them to be
in a certain ratio but you don't want
the rise and fall transition times to
change arbitrarily because that is not
the way you design a logic path in a
logic path you always have the
assumption that the transition times are
matching to maximize speed and to
minimize short circuit power to maximize
speed and minimize short circuit power
transition times in path must match this
can be seen from several references
especially logical effort so but so for
that all the logic gates in standard
cell library must be designed with a
certain ratio of the p moset and N moset
Pull up and pull down pull down
transistors so that the transition times
are matched transition to match rise
fall transition times we would the
widths of p moset and N moset must must
be in a certain ratio okay this is
important but the currents so the ratio
of these p mosit and N mosit wids in the
super threshold voltage regime that is
nominal voltage regime you may find that
you have a certain ratio which fulfills
the desired transition time ratio
suppose the desired transition time
ratio t time and fall time ratio is one
say it it may be something else also it
can be 7 it can be8 it depends on the
size of the transistor cell that is it
depends on something called track height
of the standard cell library and the
layout how how if you want smaller Gates
less area occupying Gates then you may
have some other ratio uh it need not be
one always but one is an example so in
this you to obtain this in the super
threshold voltage regime suppose you
have a certain ratio of WP by WN now in
the near threshold voltage regime the
currents are flowing through a different
mechanism in super threshold voltage
regime the currents are flowing mainly
because of drift phenomenon whereas in
near threshold or sub threshold voltage
regime the currents are mainly flowing
because of diffusion phenomenon so the
ratio of WP and WN which ensures a TR by
TF equal to 1 would not be would not
remain the same in near threshold
voltage regime okay WP by WN may not be
equal to the wp by WN of near threshold
voltage region okay so so therefore we
need to Freshly calculate freshly
estimate the wp by WN for near threshold
voltage regime so the method that
authors in this paper took is that they
matched the ion values in near threshold
voltage regime they matched the ion
values of the p moset and N mosfet and
said that that would give us the the
ratio of WP and WN which you want okay
so these are the ion values using simple
expression
though the authors in this paper they
refer to an ion value which is different
from this so but for Simplicity we can
refer to this because anyway whatever we
estimate is a back of the envelope
calculation it gives us a first estimate
you need to resort to simulations and
get to the right values because a small
change in threshold voltage because of
say d would lead to a la exponential
change in the current so therefore
whatever you calculate this way is is
only giving us a first idea so this is
this is not a very accurate idea that
you can get through these kind of
calculations how do we attempt to get a
more accurate way of doing these things
that we'll discuss in the next lecture
so this would give us a first guess for
WP by WN using this first guess obtain
WP by w n
desired using circuit simulations now
you would also have a problem that
circuit simulations would they give us
the right values because the you may not
have because of variability especially
variability so that's a question which
we'll handle in the next lecture again
so the characteristic you know is like
this isn't so somewhere here you have
threshold voltage so this is very well
characterized by sub threshold current
equation this by triode or saturation
current equation so the problem comes
about about in this regime which is near
threshold so the authors in this paper
actually
use an an empirical expression which
fits well in this near threshold reg but
the problem with any empirical
expression is that you need to First
estimate the coefficients for a given
pdk and those coefficients may also be
vulnerable to variability process
variability so therefore it is so so the
approach that we take in the course is
that we use these simple
expressions and obtain WP by WN then go
for simulations and obtain a better
value of wpw now we were discussing
inverter sizing now let us look at nand
gate sizing so nand gate we have two
cases for output discharge for charging
of the output it is almost like an
inverter rarely you will find that both
the transistors would be charging
simultaneously so we are not considering
that seriously the simultaneous
switching one of the transistors would
be switching so if one of the
transistors is switching from 1 to zero
the output would get charged but the
charging path is same as similar as the
inverter therefore if we have a
reference inverter with the value of WP
by WN as discussed previously
then what is the size of so we want to
size the P MOS and N MOS fits of the
nand gate to have a delay equal to that
of the reference inverter for output
rise fall
cases okay so now there are four cases
in this so but the output rise is almost
same similar as that of the inverter so
that's why you will have this as have a
size WP so that's not changing okay so
it would have a size which is same as
that of the p moset of the inverter what
about mosfit so now we have a series
stack of
transistors and one of the inputs we are
considering to be switching so one of
the inputs will go from 0o to one and
the other input would already be at vdd
isn't so there are two cases in this the
in the first case we are considering
that a input a is at vdd input B is
switching from Zer to vdd and in the
second case we would consider that b is
at vdd and a is switching from 0 to VD
so we will as we'll discuss further the
second case would have a larger delay so
we would size the
transistors in such a way that they
match the inverter for the second case
because that has a larger delay that is
the approach taken by this by the
authors in this paper let us look at the
discharge mechanism so now transistor B
is finding rise in input transition and
transistor a is already at vdd so at
time T is equal to Z VX is Zer because a
is already at VD and V so vgs of
transistor B is VG of transistor B is V
VB and vgs of transistor B is also VB
because VX is zero so what would happen
at time T is equal to Z when the input
starts Rising input of transistor B
starts Rising so it will the value of IB
would rise isn't the value of the
current IB would increase and I a is
initially zero at time T is equal to Z
okay but as soon as IB increases VX
would also increase because IB increases
and if I so it would charge the
parasitic capacitance at node X so that
would increase the value of VX so that
would result in a current in current I a
also because its gate voltage is at bdd
and Now drain voltage are also increased
a little a bit so that is what we see
that the value of VX also starts
increasing so at this point IIA will
also start assuming some value and which
of them so IB and I both have a non
nonzero value here so that is what we
observe and later as
you proceed further VX starts increasing
because IB is IB charges VX I a
discharges the the remaining part of IB
so therefore IB is larger I is smaller
CX is getting charged so soon the
current IIA would also start becoming
important and V out would start falling
because of this and since V out here is
also let us ske a capacitor CL and V out
here so since V out starts falling and
VX starts raising you would V out starts
fall falling and VX starts Rising you
would find that this transistor would
also enter linear regime from saturation
it had begun at saturation but it would
also enter linear regime and since its
output voltage is falling and input is
rising it would have almost a constant
current for a while and then once its
input voltage stabilizes to vdd input
voltage stabilizes to vdd it would have
a falling current this current would
start reducing
okay so when its current is constant
even then the output is getting
discharged and the what is observed is
that the charging of CX also becomes
almost zero so CX is now charged to a
certain value and it would remain at
that value for a while so and then the
transistor the output node is and at
that point what you would find is that
IIA and IB would almost become equal I
and IB would almost become equal and CX
is no not charged any further only CL is
getting discharged and after a while CL
would get quite discharged and therefore
IB the current through current IB would
also starts falling and IA would
discharge also discharge IIA would
become comparable to and slightly larger
than IB and at that time it would IIA
would charge both load capacitance and
the parasitic capacitance and all of
them would come to a 0 volts so that is
what you will observe for the lower
transistor upper transistor switching so
this is the case of upper transistor
switching now let us look at the second
case of lower transistor switching so
here this is at vdd already so initially
transistor B is at vdd so VX would have
a value which is VX of T is equal to 0
is larger than Z because VX would start
would B is at vdd and a is zero so this
is a capacitance connected to the output
node through transistor B so it would
get charged till the so the output this
parasitic capacitance node would CX
would get charged initially and it would
it would assume value which reduces the
vgs of transistor B to a value which
ensures a very small current almost zero
current through through transistor B so
once it charges to a certain value it
would not uh the node X would not charge
any further because the vgs of
transistor B is such small now that it
is not supplying any significant current
to charge CX and the leakage Cur of b
and a match so all this is for time T
smaller than zero CX charges to value
for which IB is nearly zero or IB
leakage is equal to I a leakage okay now
what happens at time T is equal to T is
larger than zero so for time T is equal
to Z or larger than or equal to Z okay
what would happen now this is Rising
once it rises to a certain value the
input of transistor a because this is
already node X is at a voltage closer to
vdd not exactly vdd so it will the out
this node X would start getting
discharged and therefore IIA Rises From
Time t equal to Z it rises further when
the value of input voltage VA is larger
than a certain value because
see the current exponential current
would increase the near threshold or sub
threshold current would increase as as
an exponential function with respect to
input voltage so therefore the increase
seems sudden the switching seems to be
sudden when the input switches the
output the current switching seems to be
sudden and here you have therefore IIA
would start increasing and IB would
start increasing only when the value of
VX is below a certain value so that vgs
of transistor B has a significant value
therefore we find that I a would
increase IB would also increase later as
the node node V's value CH reduces
significantly IIA would enter s linear
regime and IB would also enter linear
regime because V has also Fallen
somewhat later so and then uh node X is
no further charged not further charged
and the current through current IIA and
IB would match here the value at which
the current IIA and IB match is called
the value for it's called VX Target by
the authors of this paper and they
assume that the currents of IIA and IB
would match when VX reaches VX Target so
initially what we would observe for
lower transistor switching is that
initially the it is the the intermediate
node X which is discharged through IIA
and later IB also starts assuming an
important value and then it is the
output node which is which is discharged
because of the exponential nature of the
currents we observe that with increase
in Gate voltage of va the There is
almost a sudden increase in
the current I a and because VX also
starts falling with that there is a
sudden increase in the current IB so it
is not as gradual as it as it is shown
here it is quite sudden and once node X
discharges to a significant value then
the current IB also increases and you
find that node now any further reduction
in the value of and by that time VA has
reached vdd so by this time you would
find that VA has almost reached vdd so
any reduction in VX now is would result
in a reduction in IIA so that would in
result in an increase in VX because IIA
is reducing IB is having a certain value
so that would result in a an increase in
VX so let us see this initially IIA
increases IB is nearly
zero VX increases this result results in
IB increasing and IB would increase as
BX reduces and I IB would finally match
I a so CX no further increases so C VX
remains almost constant so so when the
input voltage of a reaches near
vdd and by that time if the output V out
starts discharging because VX has
reduced you would find that any so after
this any further increase so any further
so if I if VX is 2 if VX Falls since VA
is nearly equal to vdd I a would fall
but I a and IB are but I and IB are
nearly equal so VX would increase so
that's a feedback so you would find that
VX is almost constant constant once I B
is near nearly equal to I a so this
value of VX we are calling VX Target so
in this paper they call this value of VX
as VX Target or VX star when I a is
equal to IB so when I I is equal to IB
see I as VX starts falling IB starts
rising and once IA becomes equal to
IB any further reduction in VX would
result in a reduction in IIA so then it
would result in a an indefinite increase
in VX so that is not happening so once
IIA is equal to IB VX remains almost
constant because any reduction in VX
would result in a reduction in IIA and
charging of VX by IB and therefore again
a rise in VX so there is a feedback here
that's why VX remains constant and that
value of VX we call VX star in this
paper and it say that value so VX
remains at VX star and output discharges
as shown here so output discharges so
the effective value of VX for the output
switching is VX star so it is this VX
Target if if if VX star if we find its
value then we would be able to estimate
the current IB a current IB which
discharges the output node and that
current we match match with the
reference inverter discharging
current so this is the reference
inverter so we would like to estimate an
approximate value of VX Target VX star
so what are we assuming we are assuming
that VA VB is already at vdd VA has is
reached vdd VX is constant during the
discharge now at VX tar okay so if and
the currents IB is are matching IB and
IIA are matching so IB is the
discharging current for output node CL
and the output capacitance is equal in
the reference inverter and the nand gate
so then this is WP as we had discussed
this would have a value row WN so we
want to estimate that value of row so
first we would like to estimate the
value of vxar so what is the the
currents are equal so in that case I B
is equal to I a it means that the in the
current equation since we we are also
assuming that V DS of a and and B for
both the transistors is larger than 3 VT
thermal voltage therefore you can
neglect the effect of VDS so in that
case you would have the VG the effective
gate to Source voltage including DBL
effective overdrive including dabl and
body effect to be equal for both the
transistors that is VB -
vxus Lambda VG Lambda VDS what is the
value of VDS V out minus VX
plus minus gamma that is body effect V
BS so what is the value of source
voltage it is VX minus vth so this is
proportional to I similarly one can
obtain for I a also so using this we
equate that since VB is equal to vdd VA
is equal to vdd VX is equal to VX tar so
therefore you get that for IB being
equal to I a you would have vdd minus VX
Target + Lambda into V out is getting
discharged now so it has beginning to
discharge so it is again at vdd minus VX
star minus Lambda VX vdd because V out
is vdd minus VX Target minus gamma body
effect VX Target minus VTA this is equal
to this is for transistor B now this is
equal to a similar overdrive voltage for
transistor a so what is that VA is vdd
vdd minus Source voltage is is zero and
plus Lambda VDS which is VX Target minus
gamma body effect which is zero body
voltage so Source voltage is zero here
minus VT so several terms would cancel
out this is for transistor a so now you
equate both of these and you would
obtain VX star is equal to Lambda by 1 +
2 Lambda Gamma into vdd so now this is a
technology related parameter so for a
given pdk you would have this would come
out to be a constant for constant for a
pdk so that would mean that you obtain
the value of VX Target for a certain
width WN you arbitrarily choose the
value of widths of A and B keep them
equal obtain the value of VX Target
through simulations so that will give
you the value of this constant gamma
Lambda by 1 + 2 Lambda + gamma this this
is a constant for a given pdk and you
get its value so once you get this ratio
you know now this is going to be the
same for any size of w n okay so then
how do you get the value of row which
makes the discharging current equal that
of the reference inverter so for that so
you get the value of VX star now you
have obtained the value of vxar so keep
for a transistor keep the source at vxar
body voltage at vxar obtain the value of
IIA okay obtain the value of the current
and gate voltage as vdd at vdd obtain
the value of the current now this value
of current should be equal to the
reference inverter current for the same
CL if we have an equal CL here then this
current the current IB should be equal
to the reference current here so so IR
of the inverter for v in equal to vdd
and V out equal to vdd now this has a
certain width WN this has a withd WP so
what what you what are we doing so get
this value of vxar using the approach
that for a given pdk as we had discussed
earlier for that vxar obtain the value
of the current or through b through
transistor B by keeping the source
voltage and body voltage at vxar this
current is the discharging current here
and that should match the reference
current reference current obtained in
this way for WN so for that to for both
of them to match change the size or
increase the size of trans width of
transistor B and trans istor a
simultaneously keep width of transistor
b and a equal now increase that size and
increase so when you increase it to a
value row row WN for nand gate then you
get that IB matches I reference okay
that is the approach we have so in
summary what are we doing we are once we
found VX star we estimate the current
through transistor B and let that
current
equal that of the reference inverter's
discharging current irf and for that we
increase the size of transistor B
increase the width of transistor B to
row WN and then so that these currents
IB and irf match so that is how we
obtain the size of the n moset in this
now what is The Logical effort of this
now this is WN
WP WP w w n row WN and row w n and the
current carrying charging and
discharging cap capabilities are the
same so therefore logical effort is now
going to be ratio of input
capacitances so logical effort of this n
gate is therefore 1+ so to be accurate
it is w p+ row WN by WP + WN now it
depends on the ratio of WP and w
suppose the ratio of WP by WN
is8 as the authors in this paper for
their pdk they find it to be 08 if it
is8 so in this paper they find WP by
WN for
reference inverter to
be8 so if it is W if it
is8 then this would come out to be plus
row divided 1.8 so we are assuming that
capacitance of a device is proportional
to the width of the device what about
parastic delay parasitic delay delay
again with respect to that of the
inverter with respect to the parasitic
delay of the inverter so the parasitic
delay of inverter so if we assume that
this has a parasitic capacitance CP this
would also have a parasitic capacitance
CP nand and if we assume the Paras itic
capacitance to be proportional to the
width which is not a very bad assumption
because the parasitic capacitances are
mainly because of overlap capacitances
gate overlap and spacer and Junctions so
for a regular layout partic capacitance
would be almost proportional to the
width of the device so if we assume that
then the parasitic capacitance at the
output node would turn out to be row + 2
into 0
8 divided by so I'm sorry so this would
turn out to be the parasitic capacitance
at output node so for parasitic
capacitance what there are parasitic
delay what the approach that this paper
takes is that paretic delay of inverter
is RC time constant CP into
RN RC time constant of the device now
what is the value of r L of nand so this
is for inverter so what is the of nand
nand gate so since the width of the
individual device is increased by R
times so this prop this would be RN of
inverter let us call it RN I so this
would be RN I divided by row since we
have increased the width by row times so
what would be the parasitic delay it
would turn out to be and let we use El
more delay here using the
parasitic delay would be CL into 2 RN
because it has to discharge through
transistors B and A Plus CX into RN CX
needs to get discharged only through
transistor a so that is the way they
have estimated parasitic delay now we
can have a simple so taking ratio of
these two the parastic delay of nand
gate and parasitic delay of inverter if
you have some simple estimates of the
capacitances like if we assume that the
capacitances are equal to G parasitic
capacitance is equal to gate capacitance
then we get a simple value for the
parasitic DeLay So in this lecture we
discussed back of the envelope
calculation based method for
ascertaining WP by WN of inverter and of
for nand gate as an example of any gate
having stacked transistors in near
threshold voltage regime thank you
[Music]
a
[Music]
