A Digitalized Chaotic Oscillator Probabilistic Bit for Static Annealing Ising Machine
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This work presents a digital tent-map chaotic oscillator (DCO)-based probabilistic bit (p-bit) architecture for compute-in-memory probabilistic computing applications. Unlike conventional p-bits, the DCO-based p-bit takes both advantage of high throughput and robustness of pseudorandom number generator (PRNG)-based p-bits and small area consumption of the analog domain p-bits. The DCO achieves high energy efficiency of 0.041 pJ/bit. We apply this noise generator to static annealing, a schedule that maintains fixed temperature during operation, and demonstrate its effectiveness on the Max-Cut problem. Compared to conventional thermal annealing approaches that rely on gradually changing temperatures, the static annealing scheme converges faster and achieves lower final Hamiltonian values in our experiments. Our findings highlight that static annealing is not only feasible but advantageous in solving complex problems with compute-in-memory hardware.
Abstract
This work presents a digital tent-map chaotic oscillator (DCO)-based probabilistic bit (p-bit) architecture for compute-in-memory probabilistic computing applications. Unlike conventional p-bits, the DCO-based p-bit takes both advantage of high throughput and robustness of pseudorandom number generator (PRNG)-based p-bits and small area consumption of the analog domain p-bits. The DCO achieves high energy efficiency of 0.041 pJ/bit. We apply this noise generator to static annealing, a schedule that maintains fixed temperature during operation, and demonstrate its effectiveness on the Max-Cut problem. Compared to conventional thermal annealing approaches that rely on gradually changing temperatures, the static annealing scheme converges faster and achieves lower final Hamiltonian values in our experiments. Our findings highlight that static annealing is not only feasible but advantageous in solving complex problems with compute-in-memory hardware.