Propagation Delay CMOS Inverter: Transient Response V DD V DD tpHL = f(R on.C L) Rp = 0.69 R onCL V out V out CL CL Rn V in = 0 (a) Low-to-high V in = V DD (b) High-to-low Computing the Capacitances V DD V DD M1 and M2 are either in cut-off or in saturation mode during the first half (up to 50%) of the output transient. M2 V in C g4 Cdb 2 C gd 12 M4 V out Cdb 1 Cw M1 Vout 2 C g3 M3 The channel capacitance is either completely located between gate and bulk (cut-off) or gate and source (saturation). Interconnect Fanout Simplified (lumped) Model Vin The lumped model requires the gate-drain capacitance to be replaced by a capacitor to ground. So we take the Miller effect. V out CL Diffusion capacitances C db1 and Cdb2 are due to the reverse-biased pn-junction (same as we did for diode & MOS case – Ceq=KeqCj0) Thus, the only contribution to C gd12 are the overlap capacitances of both M1 and M2. − Φ0 K eq = [(Φ 0 − Vhigh )1− m − (Φ 0 − Vlow )1− m ] (Vhigh − Vlow )(1 − m) m Miller Effect Cgd 1 ∆V V out Vout ∆V Vin M1 2Cgd1 ∆V M1 Vin “A capacitor experiencing identical but opposite voltage swings at both its terminals can be replaced by a capacitor to ground, whose value is two times the original value.” C gd = 2C GD 0W ∆V CGD0 is the overlap capacitance per unit width Gate Capacitance (Cg4 , Cg3) V DD V DD Assumptions (1) M2 V in C g4 Cdb 2 C gd 12 M4 V out Cdb 1 Cw M1 Vout 2 C g3 M3 Interconnect (2) Fanout Simplified (lumped) Model Vin V out CL Ignores Miller effect on the gate-drain capacitance. This is acceptable since the connecting gate would not switch before the 50% point is reached, and Vout2 remains constant in the interval of interest. The channel capacitance of the connecting gate is constant over the interval of interest. In reality it varies from 2/3CoxWL in saturation to WLC ox in linear and cutoff. C fan−out = Cgate ( NMOS) + C gate ( PMOS ) C fan−out = (CGS 0 n + CGD 0 n + Wn Ln Cox ) + (CGS 0 p + CGD 0 p + W p Lp Cox ) Propagation Delay: First Order Analysis Integrate the capacitor (dis)charge current. tp = v2 C L (v ) ∫v i (v) dv 1 This is very difficult to solve since both CL(v) and i(v) are nonlinear functions of v. We fall back to the simplified switch model of the inverter. The voltage dependencies of the “on” resistance and the load capacitance are addressed replacing both by a constant linear element with a value averaged over the interval of interest. Req = I DSAT 1 V DD V 3 VDD 7 dV ≈ ( 1 − λVDD ) VDD / 2 VDD∫/ 2 I DSAT (1 + λV ) 4 I DSAT 9 W = k' L 2 VDSAT (VDD − VT )VDSAT − 2 CMOS Inverter Propagation Delay VDD tpHL = f(R on.CL) Ron=Req = 0.69 RonCL Vout ln(0.5) Vout CL Ron 1 VDD 0.5 0.36 Vin = V DD RonCL t t pLH = 0.69 Reqp C L tp = t pHL + t pLH 2 © Digital Integrated Circuits 2nd Reqn + Reqp = 0.69C L 2 Inverter Transient Response Hand calculated tpHL=0.69x(13Kohm/1.5)x6.1fF=36psec tpLH=0.69x(31kohm/4.5)x6.1fF=29psec Thus, tp=32.5psec Spice tpHL=39.9psec tpLH=31.7psec 3 Vin 2.5 2 (V) 1.5 out V tp = 0.69 CL (Reqn+Reqp)/2 1 tpHL tpLH 0.5 0 -0.5 0 0.5 1 1.5 t (sec) © Digital Integrated Circuits 2nd 2 2.5 x 10 -10 Inverter The previous example might give the impression that manual analysis always leads to close approximations of the actual response. This is not necessarily the case. Large deviations can often be observed between first and higher order models. The purpose of the manual analysis is to get a basic insight into the behavior of the circuit and to determine the dominant parameters. A detailed simulation is indispensable! © Digital Integrated Circuits 2nd Inverter Delay as a function of VDD (ignoring channel-length modulation) 5.5 5 4.5 t p(normalized) 4 For high VDD values (V DD>VTn+VDSATn /2), the delay is virtually independent of V DD. 3.5 3 It comes as no surprise that Req has the same behavior with V DD. 2.5 When V DD<VTn+VDSATn /2 ~ 2VTn, a sharp increase in delay is seen. This operation region should be avoided if achieving high performance is a primary goal. 2 1.5 1 0.8 1 1.2 1.4 1.6 1.8 2 2.2 2.4 V (V) DD © Digital Integrated Circuits 2nd Inverter Design for Performance q Keep capacitances small (diffusion capacitances, gate capacitances, wiring capacitance, and fanout) q Increase transistor sizes § watch out for self-loading! Once the intrinsic capacitance (diffusion capacitance starts to dominate the extrinsic load formed by wiring and fanout, increasing the gate size no longer helps reduce delay. It only makes the gate larger in area. q Increase VDD (????) – trading off performance for power. Increasing VDD above a certain values yields minimal improvement to delay. Reliability issues – oxide breakdown, hot carrier effects. © Digital Integrated Circuits 2nd Inverter NMOS/PMOS ratio 5 x 10 Making PMOS width ~ 3 times larger than NMOS width is to create an inverter with symmetrical VTC and equal tpHL, tpLH. However, this doesn’t yield minimum delay. -11 Widening PMOS improves tpLH by increasing the charging current, but degrades tpHL by causing larger parasitic capacitance. Thus, a transistor ratio must exist to optimize the delay of the inverter. tpHL tpLH t p(sec) 4.5 C L = (Cdp1 + Cdn1 ) + (C gp 2 + C gn2 ) + CW tp 4 When PMOS is β times larger than NMOS, all device capacitances scale approximately in same way; Cpd1~ β Cdn1 and Cgp2~ β Cgn2. 3.5 C L = (1 + β )(C gn1 + Cdn1 ) + CW 3 1 1.5 2 2.5 3 3.5 4 β β = Wp/Wn r= Reqp Reqn β opt = r (1 + © Digital Integrated Circuits 2nd CW ) Cdn1 + C gn2 4.5 5 R 0.69 [(1 + β )(Cdn1 + C gn2 ) + CW ](Reqn + eqp ) 2 β r t p = 0.345[(1 + β )(Cdn1 + C gn 2 ) + CW ]Reqn (1 + ) β tp = β opt = r When wiring is negligible Smaller devices yield faster designs at expense of symmetry and noise margin Inverter Device Sizing -11 3.8 The PMOS and NMOS are sized to achieve equal rise and fall delays. x 10 (for fixed load) 3.6 3.4 t p = 0.69 ReqCint ( +Cext / Cint ) = t p 0 (1 + Cext / Cint ) tp(sec) t p = 0.69 Req (Cint + Cext ) 3.2 3 2.8 Self-loading effect: Intrinsic capacitances dominate 2.6 By sizing up the inverter by S (a sizing factor to relate to a minimum sized inverter) – Cint = SCiref and Req=Rref/S. Cint consists of the diffusion + miller capacitances. 2.4 2.2 2 2 4 t p = 0.69( Rref / S )(SC iref )(+Cext / SC iref ) = 0.69 Rref Ciref (1 + 6 8 S 10 12 14 Cext ) = t p 0 (1 + Cext / SCiref ) SC iref (1) Intrinsic inverter delay tp0 is independent of the gate sizing, and is determined by the technology + layout. (2) Making S large yields high performance gain, eliminating the impact of external loads, and reducing delay of the intrinsic one – but also yield large silicon area! Cext/Cint~1.05 (Cint=3fF, Cext=3.15fF). This leads to a max performance gain of 2.05. With S=1 0 it allows us to get within 10% optimal performance, while larger device sizes only yield ignorable performance gains. By simulating, max obtainable performance improvement = 1.9 (tp0=19.3psec) © Digital Integrated Circuits 2nd Inverter Impact of Rise Time on Delay All previous assumptions are rise and fall times = 0. Only one device was assumed to be (dis)charging. In realiy, the input signal changes gradually, and temporarily PMOS and NMOS devices conduct simultaneously. This affects the current available for (dis)charging and hence delay. 0.35 tpHL (nsec) 0.3 If a latter gate is indefinitely strong, its output slope is zero, and the performance of the gate under examination is unaffected. 0.25 0.2 0.15 0 0.2 0.4 0.6 t rise (nsec) 0.8 1 The main point here is that a gate is never designed in isolation, and that its performance is affected by both the fanout and driving strength of the gate(s) feeding into its inputs. Note: It is advantageous to keep the signal rise times smaller than or equal the gate propagation delays. Aside from enhancing performance, it saves on power consumption. Keeping the rise and fall times of signals small and approximately equal is one of the major challenges in high-performance design ; slope engineering © Digital Integrated Circuits 2nd Inverter