[News] BUAA Research Team Successfully Developed Hybrid Stochastic Computing SoC Chip

As per Guangming Daily, a research team led by Professor Li Hongge from the School of Electronic and Information Engineering at Beihang University (Beijing University of Aeronautics and Astronautics, BUAA) has successfully developed a groundbreaking computing chip — the Hybrid Stochastic Computing SoC Chip. This chip, based on a fully independently developed open-source RISC-V architecture, features high fault tolerance, strong anti-interference capability, and high energy efficiency. It introduces disruptive innovations across numerical representation (redefining binary numbers), computing algorithms (in-memory computing), and heterogeneous computing architecture (SoC design). This achievement establishes a new computing paradigm based on hybrid stochastic numbers and represents an original innovation from numerical system representation to chip implementation, supporting China’s advancement in high-performance intelligent computing.

Computing power is a key metric reflecting a country’s industrial foundation. High computing power or high energy efficiency is the core of intelligent computing. However, current computing chip technologies face two major challenges: the power wall and the architectural wall. The former arises from the contradiction between the inefficiency of binary information representation and high power consumption; the latter concerns the incompatibility between non-silicon-based chips and traditional CMOS chips and their computing architectures.

To address these issues, the team has long focused on the computational mechanisms and chip design of non-binary numbers. They proposed a new type of number representation — the Hybrid Stochastic Number (HSN) — which combines binary and stochastic numbers. Professor Li explains that stochastic computing expresses values through the probability of a CMOS logic signal remaining “high” during a given time period. In other words, the frequency of high-level pulses represents the numerical probability.

This research is the first to analyze and unify the mathematical relationships among binary numbers, traditional stochastic numbers, and hybrid stochastic numbers. It constructs mathematical representations for all three systems and analyzes their characteristics, including high fault tolerance, strong interference resistance, and high energy efficiency. This work lays the groundwork for a new computing paradigm, paving the way for solving the power and architecture walls in silicon-based computing chips.

Currently, this technology is being commercialized and applied in areas like touch recognition, instrument display, and flight control computing. Professor Li states that the team is developing an extended instruction set and microarchitecture specifically for hybrid stochastic computing. The goal is to support functions including voice and image processing, intelligent computing (including large AI model acceleration), and various forms of complex computation. The on-chip dedicated operators can achieve microsecond-level latency, meeting the dual needs of dedicated hardware performance and flexible software computing.