Transcript A Robust, Fast Pulsed Flip-Flop Design

A Robust, Fast Pulsed FlipFlop Design
By:
Arunprasad Venkatraman
Rajesh Garg
Sunil Khatri
Department of Electrical and Computer Engineering,
Texas A and M University, College Station, TX
Introduction
• High speed VLSI design uses heavy pipelining
– Results in increased number of Flip-Flops
• For mobile devices
– Power consumption is the prime concern
– Requires low power Flip-Flops
– Also demand for high speed operation
• Hence there is a strong need for Flip Flops with:
–
–
–
–
High speed
Low power
Low area
Tolerance to PVT variations
Figure of Merit - Timing
• Time period
T ≥ Tcq + Tsu + d
where d – delay of the combinational circuit
Tsu – setup time of the Flip-Flop
Tcq – clock to Q delay of the Flip-Flop
• So Tcq + Tsu is the required figure of merit of the
FF, since d is circuit-dependent
Traditional Flip-Flops
• Data needs to arrive before the clock edge
– So setup time Tsu is positive
• Hence Tcq + Tsu is much higher
• Want to design a flip-flop with a goal of
minimizing the figure of merit Tcq + Tsu
• We explored different circuit designs with this
goal in mind, while ensuring that the resulting
flip-flop achieves
– Low power and area
– High speed
– Robustness to PVT variations
Pulsed Flip-Flops (P-FF)
• Broadly: consists of a pulse generator + latch
• Pulse is derived from clock edge
– So pulse is generated after clock edge
Data
D
Q
Latch
Clk
Pulse
Pulse
Clk
Generator
Clk
Pulse
Data
• Hence data can arrive even after the clock edge
(therefore Tsu may be negative)
Proposed Pulsed Flip-Flop
The proposed pulse generator design
The latch structure
Operation of Pulse Generator
•
•
•
•
Clock falls, node Z is pulled up to VDD
Clock rises, N2 discharges internal node W
Until W discharges, N2 and N1 helps to discharge Z
Very fast slew rates for falling edge of Z
Waveforms obtained at various nodes
Experimental Setup
• Implemented our Pulse Flip-Flop in BPTM 100nm
• Compared with other Flip-Flop designs
– Explicit Flip-Flop
– Improved hybrid pulsed Flip-Flop
– Traditional D Flip-Flop
• Performed Monte Carlo simulations for supply
voltage (VDD), channel length (L), threshold
voltage (VTH ) variations
– 500 Monte Carlo simulations
– 3σ = 10% for VDD, L and VTH
Pulsed Flip-Flops Compared
Explicit Pulsed Flip-Flop
Improved Hybrid latch Flip-Flop
Experimental Results
Tcq
(ps)
FLIPFLOPS
Power
(µW)
Tsu
(ps)
Thold
(ps)
Tcq + Tsu
(ps)
Clock
Load
(µm2)
µ
σ
µ
s
µ
σ
µ
σ
µ
σ
OUR
PULSED FF
95.2
8.5
8.7
1.1
-68.8
11.2
87.4
11.2
26.3
2.7
0.11
HYBRID
LATCH
PULSED FF
117
14.7
8.4
0.9
-34.4
1.9
42
4.8
82.6
11.8
0.13
EXPLICIT
PULSED FF
120.4
29.3
14.6
1.8
-54.2
4.5
108.2
11.6
65.8
7.3
0.05
Traditional
D-FF
69.9
1.5
7.6
1.3
21.4
2.5
29.9
3.1
91.2
8.7
0.09
Layout Comparison
Proposed Pulsed Flip-Flop
Master-Slave D Flip-Flop
• Our proposed pulsed Flip-Flop has 27% lesser area than a traditional
D Flip-Flop
Conclusions
• We proposed a novel pulsed Flip-Flop (P-FF) design
• The performance of our P-FF design is better than
other FFs
– 60% better Tcq + Tsu than other pulsed Flip-Flops
– 40% lower power dissipation than explicit pulsed Flip-Flop
– 27% lesser area than a master-slave D Flip-Flop
• Our P-FF is more robust to process and voltage
variations than other designs considered
• Performed Monte Carlo simulations with varying VDD, L
and VTH
• Our design has the lowest standard deviation of Tcq + Tsu
• We can further reduce area and power by sharing
pulse generator circuit between several latches