Chinese Scientists Deliver 1,000x Faster Analogue Chip
Researchers at Peking University's Institute for Artificial Intelligence, led by Dr. Sun Zhong, published on 13 October in Nature Electronics an RRAM-based analogue computing chip that reportedly solves certain matrix problems about 1,000 times faster than an H100 GPU while using over 100 times less energy. The device uses dual RRAM circuits for fast approximation and iterative refinement to reach digital-comparable precision, enabling scalable, low-power applications in 6G and AI training.
Key Points
- 1Demonstrate RRAM analogue chip achieving ~1,000× speed and 100× lower energy than H100 on matrix tasks
- 2Solve century-old precision-scalability trade-off by combining approximate and iterative RRAM circuits achieving digital-level accuracy
- 3Enable energy-efficient inference and faster LLM training, plus real-time 6G signal processing with commercial fabrication
Scoring Rationale
High novelty and peer-reviewed validation drive score, but real-world adoption scope and engineering hurdles limit immediate impact.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
