독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스주제 · LENZO AI芯片간단한 요점1 분 소요DisconfirmAI 편집부

Japanese AI Chip Startup LENZO Says A-Series Accelerator Design Complete, Targets Sample Shipments in H2 2027

LENZO states its CGLA-based A-series AI chip design is finished and is negotiating manufacturing with TSMC, requiring billions of yen to support sample shipments planned for the second half of 2027.

2026-10-19 19:00 GMT+8

Japanese AI semiconductor startup LENZO revealed at CEATEC 2026 that the design phase for its core product, the "A-series" AI accelerator, has been completed. The company is currently in discussions with foundries such as TSMC regarding manufacturing. Founded in 2024 by former Sony engineer Kenshin Fujiwara and Yasuhiko Nakashima, a professor at Nara Institute of Science and Technology (NAIST), LENZO focuses on low-power, low-latency inference technology based on its proprietary CGLA (CPU-Grounded Linear Array) architecture.

According to CEO Kenshin Fujiwara, while the team had previously validated the architecture using silicon fabricated on a 120nm process in Professor Nakashima's lab, the upcoming A-series chips are planned for advanced nodes like 4nm or 8nm. This transition requires raising billions of Japanese yen just to manufacture samples. "The only thing missing now is funding," Fujiwara stated.

The company aims to begin manufacturing as soon as capital is secured, targeting the shipment of engineering samples for the A-series in the second half of 2027. It is important to note that this timeline refers to sample delivery, not mass production or general commercial availability. At the exhibition, LENZO demonstrated a Vision-Language-Action (VLA) system combining a TokenProcessor-equipped board with a Sony camera, but this demo relied on existing prototypes rather than the final production chip's performance.

출처

로그인 / 회원가입
독서리서치레이더투자 프레임워크
eetimes.itmedia.co.jp
↗
주제

LENZO AI芯片

日本AI芯片初创公司LENZO宣布其基于CGLA架构的A系列加速器设计已完成,正与台积电洽谈制造,计划于2027年下半年出货样品。该公司由前索尼PlayStation工程师藤原健真创立,旨在通过“粗粒度线性阵列”(CGLA)架构解决传统GP…

5 단계0 건업데이트됨 2026-10-19 17:49

현재까지의 과정

  1. 2024-12-01LENZO成立,由前索尼PlayStation硬件工程师藤原健真领导,依托NAIST的CGLA架构研究。ai-market-watch.com ↗
  2. 2025-06-19发布博客文章阐述CGLA架构在AI推理和加密货币挖矿中的可重构优势。lenzo.com ↗
  3. 2026-03-13宣布完成5亿日元种子轮融资,由三菱UFJ资本领投,Incubate Fund和索尼创新基金参投。raising.fi ↗

아직 해결되지 않은 문제

  • CGLA架构在实际AI工作负载中的能效比是否真能达到宣称的比GPU低90%,是否有第三方独立基准测试数据?
  • LENZO如何突破英伟达CUDA生态的软件兼容性壁垒,以吸引企业客户而非仅依赖加密矿工?
  • 2027年下半年样品出货的具体时间表是否会因台积电产能分配或资金短缺而推迟?
전체 주제 읽기 →
리서치 계속하기 →