MemSilicon raises 400M+ yuan Series A, names ex-Graphcore China head CEO to build LPU
On October 19, MemSilicon said it closed a Series A round of more than 400 million yuan, led by Boyuan Capital with Bosch Ventures, Chuangling Capital, miHoYo, Zhongguancun Science City and GigaDevice among the participants, while existing investors including Xiaomi's strategic investment arm increased their stakes. The company said proceeds will mainly fund industrialization and commercialization of its RRAM (resistive RAM) technology and R&D of a new LPU product. It also named former Graphcore president and Greater China general manager Lu Tao as CEO and co-founder, and said it is starting first-generation LPU development for large-model inference, targeting low latency, long context and high throughput through compute-memory co-design. Those performance claims are the company's own; no independent benchmarks or shipping product were provided.
The reports show a storage-chip team spun out of academic RRAM work moving toward inference chips built around compute-memory co-design, with Lu Tao's industry experience brought in to push commercialization.
10 月 19 日，铭芯启睿（MemSilicon）宣布完成超 4 亿元 A 轮融资。本轮融资由博原资本领投，博世创投、创领资本、米哈游、中关村科学城公司、兆易创新等跟投，老股东复容投资、国开科创、恒裕资本、锦秋基金、顺禧基金、小米战投、中科创星持续加注。本轮融资将主要用于 RRAM 的产业化与商业化，以及新产品 LPU 的研发。
MemSilicon 同时宣布，前 Graphcore 总裁兼大中华区总经理卢涛加入团队，担任 CEO 和联合创始人。公司战略全面升级：在继续推进阻变存储器（RRAM）产品化的基础上，启动面向大模型推理的第一代 LPU 研发，打造「满血模型、高速 Token、越级性价比」的新一代推理引擎。
随着 Agentic AI 时代的到来，AI 落地已经从问答，转向完成复杂、长程的任务，下一步竞争的关键指标，会从峰值算力转向任务完成效率。
一次任务包含多轮推理、工具调用和上下文复用，还可能启动多个 Agent 并行工作。客户需要的，不只是更多 Token，而是在保证质量的前提下，更快地解码和输出、更多任务能并行，并降低单次任务成本。
#MemSilicon#Lu Tao#Graphcore#Boyuan Capital#Bosch Ventures#miHoYo#GigaDevice#Xiaomi#Zhongguancun Science City#RRAM#LPU