Sub-mV tunable photonic p-bits for probabilistic computing

Area of Science:

• Quantum Computing and Information Science
• Materials Science and Engineering
• Computational Science

Background:

• Randomness is increasingly recognized as a valuable resource for intelligent computation, moving beyond its traditional role as noise.
• Probabilistic bits (p-bits) offer a novel approach to computation by utilizing stochastic logic and energy-based optimization for complex problem-solving.
• Existing p-bit technologies face challenges in stable control of output probability, hindering scalability for probabilistic computing (p-computing).

Purpose of the Study:

• To introduce and characterize novel light-induced bias-tunable probabilistic-bit (LBP-bit) devices.
• To demonstrate a new framework for scalable probabilistic computing (p-computing) using LBP-bits.
• To showcase the application of LBP-bits in solving computationally intensive problems like integer factorization and Max-Cut.

Main Methods:

• Development of LBP-bit devices based on light-induced charge polarity switching in a back-to-back junction.
• Precise tuning of the stochastic bitstream probability using submillivolt bias, independent of the randomness generation mechanism.
• Implementation of a p-computing framework to evaluate performance on integer factorization and Max-Cut problems.

Main Results:

• LBP-bits successfully generate entropy through light-induced switching, providing a tunable source of randomness.
• The probability of the stochastic output is precisely controlled with minimal bias, demonstrating stable probability control.
• The proposed p-computing framework effectively solves representative problems in integer factorization and combinatorial optimization (Max-Cut).

Conclusions:

• The unique separation of randomness generation (light) and probability control (bias) in LBP-bits is crucial for scalable p-computing.
• Light serves as an effective stochastic source in these devices, enabling probabilistic computation for complex tasks.
• LBP-bits represent a significant advancement towards practical and scalable probabilistic computing architectures.