A leakage Current Replica Keeper for Dynamic Circuits
Download
Report
Transcript A leakage Current Replica Keeper for Dynamic Circuits
A Leakage Current Replica Keeper for
Dynamic Circuits
Based on the work presented in “A leakage Current Replica
Keeper for Dynamic Circuits” by:
Yolin lih, Nestoras Tzartzanis, William W. Walker
Fujitsu Laboratories of America, Sunnyvale, CA
presented in ISSCC 2006/Session 24/High performance Digital Circuits/24.4
For class presentation in Advanced VLSI Course, Tehran University, Fall 2006
By: Bardia Bozorgzadeh
Professor: Dr. S. M. Fakhraie
Outline
Introduction
Conventional keepers and their problems
Existing solutions and their problems
Leakage current replica keeper proposal
Application to a mutli-port SRAM
Measurements
Conclusion
References
Introduction
Dynamic circuits are indispensable for constructing wide highspeed OR and AND-OR gates in CMOS.
Faster and more compact than conventional logic gates.
Required to build memories.
These gates need keepers to maintain a high state during
evaluation.
Conventional Keeper
Conventional keeper is sized for worst-case (sPfN process corner)
leakage current.[1]
Also should be week enough so that a single NFET leg can pull the
dynamic node quickly in fPsN process corner.[1]
Upper limit on the number of legs decreases as processes are
scaled.[1]
Delay for conventional keeper
Keeper is sized for a max voltage drop of 0.1VDD on node DN with DC
noise of 0.15VDD applied to 1/4 of legs in sPfN corner.[1]
Gate fails to switch for more than 24 legs.[1]
Domino gates with conventional keeper will no longer be designable.[1]
Existing solutions
Conditional keeper
overhead ≥ 5 FETs
must estimate delay. (could be large)[1]
Adaptive keeper
over head ≥ 4 FETs
None of these track the two critical
process corners fPsN and sPfN. [1]
Leakage Current Replica (LCR) Keeper
Subthreshold is tracked using a replica circuit, and mirrored into the
dynamic gate keeper.
The replica current mirror can be shared with all dynamic gates
1 FET overhead/gate
Tracks all process corners as well as temperature and VDD.[1]
Only random on-die variations are not tracked, but after margining for
these variations, LCR still has an advantage.[1]
Sizing LCR Keeper
Wnprl = sf.∑Wni
With sf=1 will hold DN only to same voltage as
KPR.
V(KPR) design critical[1]:
too low : poor replica
too high : ΔVtp sensitive
sf is selected to be approximately 10, then p1 is
driven to triode region to keep DN close to VDD.
Choose long L for p1 and p3 to reduce Vt
variations.[1]
The size of p2 isn’t critical as long as it is bigger
than p1. [1]
Selecting a safety factor
Delay Comparison
LCR Keeper is usable for up to 32 legs.
sPfN process corner presents the worst-case delay for more than 20 legs
due to replication of the leakage current by the keeper.[1]
With 16 to 24 legs LCR is 25 to 40% faster than conventional keeper. [1]
Application to multi-port SRAM
1024 word × 72 bit
3 write / 4 read
3 stage domino read
path
Fabricated in 90nm,
1.2V CMOS
1.34mm × 1.31mm
sf 6 to 10
The 1024 × 72 3W/4R SRAM block diagram [1]
LCR SRAM single-ended 3-stage
domino read path
Cont’d
Previous SRAMs in this technology are desinged using high-Vt NFETs in
the dynamic gates but using LCR keeper allowed us to switch to low-Vt
NFETs. [1]
Since all read paths are identical only one current mirror is used for the
entire SRAM.[1]
2 × minimum length channel is used for p1 and p3 to increase Vtp and
reduce its variations.
Simulated access time comparison
The delay from clk to RWD is identical for both circuits because they use
identical static logic.[1]
Conclusion
A leakage current replica keeper is proposed to improve
scaling of dynamic gates.
Overhead is 1-FET per gate plus a portion of the replica
circuit.
For equal noise margins, either:
more legs are possible;
gate is faster with the same number of gates.
A fairly large safety factor is needed to account for random ondie process variation especially FET Vt variation.
References
[1] Yolin lih, Nestoras Tzartzanis, William W. Walker. “A leakage Current Replica
Keeper for Dynamic Circuits” ISSCC 2006/Session 24/High performance Digital
Circuits/24.4
[2] N.H.E. Weste and David Harris, “CMOS VLSI Design. A Circuit and System
perspective”. Third Edition, Boston: Addison-Wesley, 2005.
[3] A.Alvandpour et al., “A Conditional Technique for Sub-0.13µm wide Dynamic
Gates,” Symp. VLSI Circuits, pp. 29-30, 2001
[4] C. R. Gautheir et al., US Patents #6,914,452 and #6,894,528, 2002.
Thank You