230 likes | 312 Views
Energy Optimization of Probabilistic BooleaN Circuits. Yung-Chun Hu & Ching -Yi Huang & Black 2013/12/02. Outline. Motivation Introduction Problem Formulation Probabilistic gate assignment and selection Optimization: Near-redundant removal Issues Future work. Motivation.
E N D
Energy Optimization of Probabilistic BooleaN Circuits Yung-Chun Hu & Ching-Yi Huang & Black 2013/12/02
Outline • Motivation • Introduction • Problem Formulation • Probabilistic gate assignment and selection • Optimization: Near-redundant removal • Issues • Future work
Motivation Moore’s Law v.s. power issue
Solutions • Multiple Processors • 3D IC • FinFET • Single Electron Transistor(SET) • Probabilistic CMOS(PCMOS) • …
Probabilistic CMOS Lower Vdd Power Voltage
Problem formulation • Given • A deterministic circuit • Correctness constraint • Derive • An energy optimized probabilistic Boolean circuit
Outline • Motivation • Introduction • Problem Formulation • Probabilistic gate assignment and selection • Optimization: Near-redundant removal • Future work
Probability selection • PTM 45nm bulk • Vtn 0.466 ; Vtp 0.412 • Wn 0.415 ; Wp 0.63; L 0.5 Energy (normalized) Probability
P-gate assignment • Observation • Same input pattern and stuck-at fault, different outputs 1 A X 1 S-A-0 B 1/0 Y 1 F Z 1 C D 1 S-A-0 A X 1 1 B Y 1 F Z 1 C D
P-gate assignment • We usually use testability/observability to measure the difficulty of propagating fault effects to POs • Fault effects caused by probabilistic behavior have greater chance to be blocked if its testability is lower
P-gate assignment Random Testability-order
Outline • Motivation • Introduction • Problem Formulation • Probabilistic gate assignment and selection • Optimization: Near-redundant removal • Future work
Redundant Removal • In deterministic circuit, we can inject SA faults and remove redundant wires. SA fault Observable? No Remove the wire
Redundant Removal • Example The only possible test pattern is (1,1,0) S-A-0 1 A 1 B 0 C F
Redundant Removal • Example The only possible test pattern is (1,1,0) S-A-0 1 A 1 B 0 C 1/0 1 F 1
Redundant Removal • Example 0 0 C 1/0 1 F 1
Near-redundant removal • Regard signals which have high probability to be 0 or 1 as stuck-at-faults E A 0.0375 w 0.9 0.9625 0.3 B 0.075 Correctness=0.99375 0.25 C D Assume input space is a full set, w has probability of 0.9625 to be 1
Near-redundant removal • In PBC, we may regard signals which have high probability to be 0 or 1 as stuck-at-faults. 0 Correctness=0.977 The node reduction is 66%.
Probability selection issue • Number of probability? The more probability we use, the more voltages we need • Realization Power island Power islands in an SoC Power islands in a circuit? 1.0V 1.1V Scale down 0.8V 0.9V
Power Issue P-gate→P-gate General-gate→P-gate power voltage
Power Issue P-gate→General-gate power voltage
Future work • Observability-Probability-Correctness observation