Impact of CMOS Technologies scaling in Leakage Reduction Techniques - - PowerPoint PPT Presentation

Impact of CMOS Technologies scaling in Leakage Reduction Techniques

Saad Arrabi, Taeyoung Kim

OUTLINE

• INTRODUCTION
• BACKGROUND
• METHODOLOGIES
• SIMULATION RESULTS
• CONTRIBUTION
• FUTURE WORK

INTRODUCTION

• MOTIVATION

– Power consumption is the major concern in CMOS semiconductor industries. Leakage is one of the big part of power consumption. – Off current leakage is being more serious problem as the technology size shrinks.

• GOALS

– Explore what conventional techniques for leakage reduction is used. – Investigate how effectively they can cope with new technology scaling.

BACKGROUND

Sleep Stack Zigzag CMOS Sleepy Stack

Sleep

• NMOS Sleep transistor usually used

– To reduce resistance of “on” status

• Pros

– Reduction dynamic and leakage power

• Cons

– Reduce performance and requires additional area – Energy is consumed by charging and discharging the virtual rails

• Knob

– Footer vs. Header – Width – Sharing vs. Non-sharing Sleep TR

• Sharing will decrease the delay

STACK

• Doesn’t need to go to sleep mode
• Introduces new capacitance and delay
• Knobs

– Width: more width, more leakage, more area, less delay – Placement: Headers only, some inputs.

• Different leakage reduction and less area

Methodologies

• RANDOM CIRCUIT

– Simulate glitching – Large enough to be between 2 registers stages – Not fully accurate, but good enough

• DELAY

– Dynamic Delay, 50% of input to 50% of output – Wake up Delay, 50% of input signal to 90% of rail voltage

• ENERGY

– Dynamic Energy, averaged between 0->1, 1->0 – Leakage Energy, averaged between static 0, 1 – Wake up Energy, same as delay

Dynamic Power vs Total Widths

1 2 3 4 5 0.5 1 1.5 2 2.5 3 Total Width (x Minimun) Dynamic Power (Nomalized by BASECASE) 22nm Shared Sleep 32nm Shared Sleep 45nm Shared Sleep 65nm Shared Sleep 22nm Non-shared Sleep 32nm Non-shared Sleep 45nm Non-shared Sleep 65nm Non-shared Sleep 22nm Stack 32nm Stack 45nm Stack 65nm Stack

Dynamic Delay vs. Total Widths

1 2 3 4 5 1 2 3 4 5 6 7 Total Width (x Minimum) Delay (Normalized by BASECASE) 22nm Shared Sleep 32nm Shared Sleep 45nm Shared Sleep 65nm Shared Sleep 22nm Non-shared Sleep 32nm Non-shared Sleep 45nm Non-shared Sleep 65nm Non-shared Sleep 22nm Stack 32nm Stack 45nm Stack 65nm Stack

Leakage reduction variation

• 2

2 4 6 8 0.00 0.05 0.10 0.15 0.20 0.25 0.30 Occurrences Leakage reduction normalized

22 nm leakage variation

share noshare stack

• 5

5 10 0.00 0.50 1.00 1.50 2.00 Occurrences Leakage reduction normalized

65 nm leakage variation

share noshare stack

Comparison

Metric Stack Sleep Added delay High Medium Added energy High Low Leakage in active mode Medium None Leakage in sleep Medium High Complexity Simple Complex Area Big area Medium area

Contribution

• Showed the effectiveness of some techniques

for future technologies

• Provide a base work for knob tweaking for

leakage reduction techniques

• Provided pareto curve methodology for

leakage and delay

Future Work

• Analysis of the other techniques using more

general circuit, such as full adder and SRAM.

• Analysis of different set of knobs like

placement location and voltage threshold.

• Evaluation of new scaling technologies (16nm)
• r using other Predictive Technology Model.
• Use different cost formulas like complexity