Dynamic Logic Circuits
Dynamic Logic Circuits
Dynamic Logic Circuits
9-1
Pass transistor circuits Voltage bootstrapping Synchronous dynamic circuit techniques Dynamic CMOS circuit techniques High-performance dynamic CMOS circuits
9-2
9.1 Introduction
Static v.s. Dynamic Static Logic Gates
Valid logic levels are steady-state operating points Outputs are generated in response to input voltage levels after a certain time delay, and it can preserve its output levels as long as there is power. All gate output nodes have a conducting path to VDD or GND, except when input changes are occurring.
9-4
Operation
CK
CK = H, D=H or L : CX is charged up or down through MP, and X becomes H or L (depends on D input) since MP is on D and X are connected. CK = L: X is unchanged since MP is off and CX is isolated from D, and the charge is stored on capacitances CX. For X = H, Q = L and Q = H For X = L, Q = H and Q = L
Vin=VDD D
S MP
ID X
Vx Cx
Fig. 9.1
CK
CK
VGS = VDD - VX, VDS = VDD - VX = VGS. Therefore, VDS> VGS VT,MP MP is in saturation.
CX
Fig. 9.2
2 dVX = kn (V DD V X VT ,MP) 2 is subject to substrate bias effect and Note that the VT,MP dt
therefore, depends on the voltage level VX. We will neglect the substrate bias effect for simplicity.
9-7
Integrating the above equation with t from 0 t and VX from 0 VX, we have
2C dt = X kn 0
t VX
(V
0
dVX 2 DD V X V T ,MP)
VX
Therefore,
1 = 2CX kn V DD V X VT ,MP 0
1 1 t = 2CX kn V V V V V DD X T , MP DD T , MP
and,
k n (V DD V T ,MP ) t 2C X V X (t ) = (V DD V T ,MP ) ( ) 1 + k n V DD V T ,MP t 2C X
9-8
0 approaches a limit value V VX rises from 0V and max = VX(t)|t= = VDD-VT,MP, but it can not exceed this value, since the pass transistor will turn off at this point (VGS=VT,MP). Therefore, it transfers a weak logic 1. The actual Vmax by taking the body effect into account is, V max = V DD V T 0,MP 2 F + V max 2 F and tcharge = time to VX = 0.9Vmax,
Fig. 9.3
t
9 - 9 Body
charge
9 - 10
9 - 11
Basic Principles of Pass Transistor Circuits: Logic 0 Transfer Logic 0 Transfer: VX(t=0)=Vmax= VDD VT,MP, Vin=VOL=0V, CK= 0 VDD
Soft note X Cx CK Vin MP Vx
Vin=0
S MP
ID X
Vx Cx
CK
VGS = VDD, VDS = Vmax = VDD VT,MP. Therefore, VDSVGS VT,MP MP is in linear region.
Fig. 9.6
CX
dVX kn [ ( = 2 V DD VT ,MP)V X V 2 X] dt 2
Note that the VSB=0. Hence, there is no body effect for MP (VT,MP= VT0,MP). But the initial condition VX(t=0)=VDD VT,MP contains the threshold voltage with body effect. To simplify the expressions, we will use VT,MP in the following.
9 - 12
dt =
0
Therefore,
t=
2 (V DD V T , MP ) V X CX ln k n (V DD V T , MP ) VX
and,
(t) =
2 (V 1+ e
tk n (V
DD
V
DD
T , MP
T , MP
)/ C X
9 - 13
where,
(2 0.9 )(V DD V T ,MP ) CX ln t 90% = k n (V DD V T ,MP ) 0.9 (V DD V T , MP ) = t10% = CX ln (1.22 ) k n (V DD V T ,MP ) k n (V DD V T ,MP ) CX 1.9 ln 0.1
Fig. 9.7
9 - 14
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage
At t = 0, CK=0, VX= Vmax, Vin =0. The charge stored in CX will gradually leak away, primarily due to the leakage currents associated with the pass transistor. The gate current of the inverter driver transistor is negligible.
Vin =0 MP CK=0
Ileakage Vx
Igate=0
Cx
Fig. 9.8
Ileakage VX CX
Isubthreshold Ireverse
n+
Fig. 9.9
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage (Cont.)
VCK=0 Vin=0 n+
p-type Si
Ileakage VX n+ CX
Vx Cj(VX)
Isubthreshold
Ireverse
Drain-substrate pn-junction
Fig. 9.10
Isubthreshold is the subthreshold current for the pass transistor with CK=0. Ireverse is the reverse current for the source/drain pn junction at node X Cj (VX) : due to the reverse biased drain-substrate junction, a function of VX Cin: due to oxide-related parasitics, can be considered constants.
9 - 16
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage (Cont.)
Ileakage Isubthreshold
Cj
Ireverse
Drain-substrate pn-junction
The total charge stored in the soft node can be expressed as, Q = Qj (VX) + Qin where Qin = CinVX The total leakage current can be expressed as the time derivative of the total soft-node charge Q
I leakage = dQ dt dQ j (V X ) dQ in = + dt dt dQ j (V X ) dV X = + C in dV X dt dt dV X
9 - 17
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage (Cont.)
Where
dQ j (V X ) dV X = C j (V X ) = AC j 0 AC j 0 SW + V 1+ X 1+ V X
0 SW
0 =
kT ND N A ln q ni2
0SW =
kT ND N ASW ln q ni2
Therefore,
AC j 0 PC j 0 SW + + C in dV X I leakage = dt V V 1+ X 1+ X 0 0 SW
We have to solve the above differential equation to estimate the actual charge leakage time from the soft node.
9 - 18
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage (Cont.) A quick estimate of the worst-case leakage behavior
Assume that the minimum combined soft-node capacitance is CX,min = Cgb + Cpoly + Cmental + Cdb,min Cdb,min is the minimum junction capacitance, obtained when VX=Vmax The worst-case holding time (thold) is the shortest time for VX to drop from its initial logic-high value to the logic threshold voltage due to leakage. thold = Qcritical,min/Ileakage,max where Vth
Qcritical,min =CX,min (Vmax-VDD/2)
9 - 19
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage (Cont.)
Example 9.2: Consider the soft-node structure shown below, which consists of the drain (or source, depending on current direction) terminal of the pass transistor, connected to the polysilicon gate of an nMOS driver transistor via a metal interconnect. Question: is to estimate thold if VDD=5V and the soft-node is initially charged to Vmax.
MP CK 3 1 MP 4 1 CK
diffusion
Vx Cx
M1 soft node 6 5 5 2 3
metal polysilicon
6 2 M1 4 1 2
9 - 20
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage (Cont.)
Material parameters: VTO = 0.8V = 0.4V1/2 |2F| = 0.6V 0 = 0.88V 0SW = 0.95V Ileakage,max = 0.85 pA COX = 0.065 fF/m2 Cmetal = 0.036 fF/ m2 Cpoly = 0.055 fF/ m2 Cj0 = 0.095 fF/ m2 Cj0SW = 0.2 fF/m
3 1 MP 4 1 CK 6 5 5 6 2 M1 4 1 2
2 diffusion metalpolysilicon 3
Oxide-related (constant) parasitic capacitances Cgb = COXWLmask = 0.065 fF/m2 (4 m2 m) = 0.52 fF Cmetal = CmetalWLmetal = 0.036 fF/m2 (5 m5 m) = 0.90 fF Cploy = CpolyWLpoly = 0.055 fF/m2 (36+6+2 m2) = 2.42 fF
9 - 21
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage (Cont.)
Parasitic junction capacitance By zero-bias unit capacitance values in the previous slide, we have Cbottom = AbottomCj0 = 0.095 fF/m2 (36 m2 + 12 m2 ) = 4.56 fF Csidewall = Cj0SWPsidewall = 0.2 fF/m2 (30 m) = 6.00 fF Therefore Cdb,max = Cbottom + Csidewall = 4.56 fF + 6.00 fF = 10.56 fF The minimum drain junction capacitance is achieved as the junction is biased with Vmax. We need to find Vmax to determine Cdb,min Vmax = 5.0 - 8.0 - 0.4 ( 0.6+ Vmax - 0.6 ) C bottom C sidewall + C db,min = Vmax = 3.68 V V 1 + X ,max 1 + V X ,max 0 0 SW Therefore,
= 4.56 6.0 + = 4.71 fF 3.68 3.68 1+ 1+ 0.88 0.95
9 - 22
Basic Principles of Pass Transistor Circuits Charge Storage and Charge Leakage (Cont.)
Combining the Oxide-related (constant) parasitic capacitances with the parasitic junction capacitance, CX,min can be got as CX,min = Cgb + Cpoly + Cmental + Cdb,min = 0.52 + 2.42 + 0.90 +4.71 = 8.55 fF The amount of the critical charge drop is Qcritical = CX,min(VX,min-VDD/2)=8.55 (3.682.5)=10.09 fC Finally, thold = Qcritical /Ileakage,max=11.87ms The worst-case hold time for this structure is relatively long, even with a very small soft-node capacitance of 8.55fF. It means that the logic gate can be preserved in a soft node for a long time period when the leakage current is small.
9 - 23
The Voltage bootstrapping is a technique to overcome the threshold voltage drops of the output voltage levels in pass transistor gates or enhancement-load inverters and logic gates. Consider the following circuit with VXVDD M2 is in saturation. If Vin is low, the maximum output voltage is limited as Vout(max) = VX VT2(Vout)
VDD Vx M2 Vout Vin M1 Cout
Fig. 9.11
9 - 24
Fig. 9.12
Initially, let Vin=H M1 and M2 are on, and Vout=L. Now Vin goes to L M1 turns off, and Vout starts to rise. This change will be coupled to VX through the bootstrap capacitor, Cboot.
9 - 25
X DD
V T 3
dV
C boot C S + C boot
DD
DD OL
dV
out
= (V
V T 3) +
If Cboot >> CS, then for Vout rising to VDD, VX(max) 2VDD VT3 VOL > VDD VT2. for realistic values of the voltages. Thus, it is feasible to use the circuit to obtain Vout =VDD.
C boot (V C S + C boot
DD
V OL )
9 - 26
VT 2 Vout=VDD +VT 3 VX Cboot = CS + Cboot V DD VOL . VT 2 Vout=VDD +VT 3 VX Cboot = CS V DD VOL VT 2 Vout=VDD VT 3 VX
CS is the sum of the parasitic source-to-substrate capacitance of M3 and the gate-to-substrate capacitance of M2.
9 - 27
Fig. 9.13
1 1 2
1
t
Fig. 9.14
phase1
phase2
Fig. 9.15
Logic levels are stored on input capacitances during the inactive 9 - 29 clock phase.
Vout
Vin
Cin1
Cout1
Cin2
Cout2
Cin3
Cout3
Fig. 9.16
9 - 30
9 - 31
9 - 32
Instead of biasing load transistors with a constant gate voltage, a clock signal is applied to the gate of the load transistor power dissipation and silicon area are reduced. The power supply current flows only when the load devices are activated by the clock signal, the power consumption is lower than the depletion-load nMOS logic.
1
VDD
VDD
VDD
2
Vout
Vin
Cin1
Cout1
Cin2
Cout2
Cin3
Cout3
Fig. 9.19
9 - 33
1
A B C nMOS Logic Stage 1 nMOS Logic Stage 2
1
Z
Fig. 9.20
General Circuit Structure of Ratioed Synchronous Dynamic Circuit
9 - 34
9 - 35
2
Vout1
VDD
1
Vout2
VDD
2
Vout3 Cin3 VDD Cout3
Vin
Cin1 VDD
Cout1
Cin2
2=H
2
Vout1
2
Vout3 Cin3 Vout3VOL Cout3
Vin
Cin1
Cout1 Vout1VOL
Cin2
Cout2
VOL kdriver/kload Ratioed Dynamic Logic. Cout1, Cin2 & Cout2, Cin3 interact Charge Sharing
9 - 36
VDD
VDD
9 - 37
Fig. 9.21
2
Z
Fig. 9.22
General Circuit Structure of Ratioless Synchronous Dynamic Circuit
9 - 38
VDD
1
Vout2 0
VDD
Cin3
2=H
VDD Vout1
VDD
Vin
VOH Cin1
VDD
1
Vout2 0
VDD
Charge Sharing Cini << Couti-1 for i=2,3 Minimum Charge Sharing
9 - 40
Vb=VOH
Vi1=0 Cin1
Vb=VOL 0
Vin1=1 Cin1 Cout1
Va Cin2
Charge Sharing
2 = 0: Qout2 = Cout2Vb and Qin3 = Cin3Va 2 = 1: Qtotal = Cout2Vb + Cin3Va and Ctotal = Cout2 + Cin3 The resulting voltage across Ctotal is VR = Qtotal / Ctotal = (Cout2Vb + Cin3Va )/ (Cout2 + Cin3) If Vb = VOH and Va << Vb VR Cout2VOH /(Cout2 + Cin3) VR VOH if Cin3 << Cout2
9 - 41
A B C
9 - 42
F1 Stage 1 D
1 2
Stage 2
1
Fig. 9.23
VDD
Vin
CK 9 - 43
VX CX
The single-phase CMOS shift register is built by Cascading identical inverter units Driving each stage alternately with the CK and CK. Ideally: The odd-numbered stages are on as CK=1, while the evennumbered stages are off the cascaded inverter stages are alternately isolated. Practically: The CK and CK are not a truly nonoverlapping signal pair, since their waveforms have finite rise and fall times. One of the signals is generated by inverting the other the clock skew is unavoidable. True two-phase clocking is preferred over single-phase clocking.
CK CK CK
V1
CK
V2
CK
V3
CK
V4
9 - 44
Fig. 9.225
Me
=0 Mp on and Me off C precharges to VDD (output is not available during precharge) =1 Mp off and Me on C selectively discharges to 0 (output is only available after discharge is complete)
precharge
evaluate precharge t t
Vout
9 - 45
Mp Vout
A1 B1 A2 B2 A3
Me
Fig. 9.26
Advantages
Need only N+2 transistors to implement a N-input gate. Low static power dissipation No DC current paths to place constraints on device sizing Input capacitance is same as pseudo nMOS gate. Pull-up time is improved by active switch to VDD.
Disadvantages
The available time of output is less than 50 % of the time. Pull-down time is degraded due to series active switch to 0. Logic output value can be degraded due to charge sharing with other gate capacitances connected to the output. Minimum clock rate determined by leakage on C. Maximum clock rate determined by circuit delays. Input can only change during the precharge phase. Inputs must be stable during evaluation; otherwise an incorrect value on an input could erroneously discharge the output node. (single phase P-E logic gates can not be cascaded) Outputs must be stored during precharge, if they are required during the next evaluate phase.
9 - 47
Mp2
Vout2
precharge evaluate t Vout1 does not switch from 1 to 0 fast enough t correct state erroneous state t
2nd 1 Me2
inputs
Vout Vout
Evaluate: Me1, Me2 ON Mp1, Me2 OFF Problem: All stages must evaluate simultaneously one clock does not permit pipelining of stages.
9 - 48
Static inverter serves to buffer the logic part of the circuit from its output load
Vout
X nMOS Logic
inputs
precharge evaluate
1 t 9 - 49
=0 X precharges to VDD, and Vout = 0. =1 X remains high, and Vout remains low. X discharges to 0, and Vout changes from 0 to 1. Fig. 9.27
9 - 50
9 - 51
9 - 52
9 - 53
VDD
VDD
X3 nMOS Logic
inputs
nMOS Logic
X1 X2 X3
evaluate
precharge
evaluate teval t t t t
9 - 54
The problem in cascading conventional dynamic CMOS occurs when one or more inputs make a 1 to 0 transition during evaluation. Domino circuits can fix the above problem During the evaluation, each buffer output can make at most one transition (from 0 to 1), and thus each input of all subsequent logic stages can also make at most one (0 to 1) 9 - 55 transition.
9 - 56
Vout
Assume that all inputs are low initially, and the voltage across C2=0V During the precharge, C1 is charged to VDD If transistor N switches from 0 to 1 during the evaluation phase, the charge initially stored in C1 will be shared by C2. Therefore, the value of VX will reduced.
9 - 57
VX nMOS Logic
inputs
Push VX to VDD unless there is a strong pull-down path between Vout and ground
Fig. 9.33
9 - 58
Use separate pMOS transistors to precharge all intermediate nodes Vout1 in nMOS pull-down tree which have a large parasitic capacitance. Effectively eliminate all charge sharing problems during evaluation Vout2 Allow implementation of multipleoutput domino structures. Can cause additional delay since the nMOS tree need to drain a larger charge to pull down VX
9 - 59
Another Way: Use a smaller threshold voltage the final stage output is not affected by lowering of VX trade off the pull-up speed (weaker pMOS transistor) Fig. 9.34
9 - 60
Let C1 = C2 = 0.05pF. VX1 = 0, and VX2 = 0 at t=0 Without this extra pMOS transistor Precharge: VX1 VX2 Evaluation: VX1 = VDDC1/(C1+C2) = VDD/2 With this extra pMOS transistor Presharge: VX1 = VX2 Evaluation: VX1 = VDD See pp.392~393 for the HSPICE simulation result Note that there is a speed penalty for adding this extra pMOS precharge transistor.
C4 P4 P3 P2 P1 C0 G1 G2 G3 G4 C3 C2 C1
Gi = Ai Bi Pi = Ai Bi
Fig. 9.35
A B C D
Mp Vout
CL
R0
Me
0 C0
R1 1 C1 CL
9 - 62
Fig. 9.36
R0
R1 1
nMOS Logic
pMOS Logic
nMOS Logic
to nMOS stage to pMOS stage nMOS stage all stages all stages nMOS stage precharge evaluate evaluate precharge pMOS stage pMOS stage pre-discharge pre-discharge
Advantages
Fig. 9.37
9 - 64
An Inverter is not required at the output of stages Allow pipelined system architecture Disadvantages: Also suffer from charge sharing and leakage
Fig. 9.38
=L: nMOS precharges to H, and pMOS predischarges to L. =LH: All cascaded nMOS and pMOS logic stages evaluate one after the other.
9 - 65
NORA CMOS Logic (NP-Domino Logic) Examples (Cont.) Pipelined System Architecture: See Fig. 9.39 Use of CMOS2 latches (three state latches storing on logic inputs.) Zipper Logic: See Fig. 9.40 Identical to NORA except for weird clock signals that keep precharge devices weakly on to handle charge leakage and charge sharing
9 - 66
9 - 67
9 - 68
nMOS Logic
pMOS Logic
to next N-block
N-block
P-block
Fig. 9.41
Using tristate inverters between stages decouples the stages and enables pipelined operation
9 - 69
=L: nMOS blocks precharge to VDD pMOS blocks evaluate by selective pull-up to VDD =H: pMOS blocks pre-discharge to VDD nMOS blocks evaluate by selective pull-down to 0V is not used, no clock skew problem can arise. Provide similar performance to NORA structure
Fig. 9.42
9 - 70
Need only 11 transistors. Static Edge Triggered D Flip-flop (see Fig. 8.30) need 16 transistors. Common Advantages of dynamic Logic Styles Smaller area than fully static gates. higher speed: smaller parasitic capacitances. Glitch free operation if design carefully
9 - 71
9 - 72
Summary
Full complementary static logic is best option in the majority of CMOS circuits. Noise-immunity is not sensitive to kn/kp Does not involve precharge of nodes Dissipate no DC power Layout can be automated Large fan-in gates lead to complex circuit structures (2N transistors) Larger parasitics Slower and higher dynamic power dissipation than alternatives No clock
9 - 73
Summary (Cont.) Pseudo-nMOS static logic finds widest utility in large fan-in NOR gates. Require only N+1 transistors for N fan-in Smaller parasitics Faster and lower dynamic power dissipation than full CMOS Noise immunity sensitive to kn/kp Dissipate DC power when pulled down Not well suited for automated layout No clock
9 - 74
Summary (Cont.)
CMOS domino logic should be used for low-power, high speed applications Require only N+k transistors for N fan-in, size advantages of pseudo-nMOS. Dissipate no DC power Noise immunity is not sensitive to kn/kp Use of clocks enables synchronous operation Rely on storage on soft node Require exhaustive simulation at all the process corners to insure proper operation Some of the speed advantage over static gates is diminished by the required per-charge (predischarge) time.
9 - 75