突破世纪难题！我国成功研制

Core Insights - A new type of chip has been successfully developed in China, which is based on resistive random-access memory (ReRAM) and achieves high precision and scalability in analog matrix computation, rivaling digital computation in accuracy [1][2] Group 1: Technological Breakthrough - The research team from Peking University has created a high-precision, scalable analog matrix computing chip that can solve large-scale MIMO signal detection problems with a computational throughput and energy efficiency that surpasses current top digital processors (GPUs) by 100 to 1000 times [1] - The chip achieves a precision of 24-bit fixed-point accuracy, addressing a century-old challenge in the scientific community regarding the integration of high precision and scalability in analog computing [1][2] Group 2: Performance Metrics - The team successfully demonstrated the ability to invert a 16×16 matrix with a relative error as low as 10⁻⁷ after 10 iterations [2] - For a 32×32 matrix inversion problem, the chip's computational power exceeds that of high-end GPUs, and for a 128×128 matrix, its throughput is over 1000 times that of top digital processors, completing tasks in one minute that would take traditional GPUs a day [2] - The energy efficiency of this chip is more than 100 times better than that of traditional digital processors at the same precision level, providing critical technological support for high-efficiency computing centers [2]