- Funding backs commercialization of next-generation AI processors built on computing-in-memory architecture
- The financing highlights sustained investor interest in chip designs aimed at overcoming AI’s power and memory bottlenecks
Yizhu Tech (亿铸科技), a Suzhou-based AI processor startup developing computing-in-memory chips, has raised more than 1 billion yuan ($147 million) in a new financing round as investors continue to back alternative semiconductor architectures designed for large-scale artificial intelligence workloads.
The round, announced on July 2, drew a diversified group of undisclosed investors, including leading early-stage technology funds, local state-backed investors, industry customers, overseas institutions and corporate venture capital arms of listed companies.
The company has now raised more than 1.5 billion yuan since its founding, with existing shareholders including CAS Star, Legend Star, ABC International and Xction Capital.
Founded in 2020, Yizhu Tech describes itself as the world’s first company to develop high-performance AI processors based on a computing-in-memory architecture.
Its chips target data centers, cloud computing, autonomous driving and edge computing applications, and the company has been recognized as one of China’s national-level potential unicorns.
Embedding computation into memory
Unlike conventional chips based on the von Neumann architecture, where memory and computation are physically separated, computing-in-memory integrates computation directly into memory arrays.
The approach is intended to reduce the massive energy and latency associated with moving data between memory and processors.
This represents a growing bottleneck for training and running large AI models. Industry estimates suggest data movement accounts for 60% to 80% of total chip power consumption.
Proof of concept
Yizhu Tech combines this architecture with ReRAM (resistive random-access memory), a next-generation non-volatile memory technology, while adopting a fully digital chip design.
In 2023, the company introduced what it calls a “super-heterogeneous computing-in-memory architecture.” It has since powered on a proof-of-concept AI processor based on ReRAM, with third-party testing showing energy efficiency exceeding that of conventional AI chips by more than tenfold, according to the company.
Same bet, varying pathways
Yizhu Tech joins a growing group of Chinese startups pursuing computing-in-memory technologies, including Nano-Core Chip (微纳核芯), Witmem Technology (知存科技), Houmo.AI (后摩智能), Reexen Technology (九天睿芯) and Light Standard (光本位), each pursuing different technical approaches and commercialization strategies.
