Physical AI News (Japan)

Physical AI, robotics and compute — reported from Japan.

日本語
· Research & technology· 日本語で読む

Hitachi Develops AI Platform Technology Reportedly Cutting Chiplet Design Candidates by More Than 90%

ITmedia reported that Hitachi developed AI platform technology to support chiplet package design. According to the report, agentic AI narrows design candidates by more than 90%. The technology was reportedly validated in package design using three types of chiplets.

What happened

On November 16, 2026, ITmedia reported that Hitachi developed AI platform technology to support chiplet package design. According to the report, agentic AI narrows design candidates by more than 90%, and the technology was validated in package design using three types of chiplets.

Why it matters

If ITmedia’s report is accurate, the technology could support the narrowing of candidates during the design process for considering chiplet combinations and package structures. The reported reduction of more than 90% refers to the number of candidates; it does not indicate improvements in design time or product performance, making the conditions of application and validation through measurements important.

Analysis

The main reported result is a reduction of more than 90% in candidates at the design stage, when chiplet combinations and placement, as well as package structure, are determined. This should be positioned as technology that supports candidate exploration, not as a production technology that automatically finalizes design proposals. The report says that electrical characteristics, communications and heat need to be evaluated. If the candidate-narrowing system is put to practical use, it could allow designers to limit the options subjected to detailed evaluation. However, the candidate reduction rate alone does not show how much design time was shortened or how the performance and reliability of the selected design changed. Validation in package design using three types of chiplets indicates that, at least according to the report, the evaluation was limited to a specific configuration. Whether candidates can be narrowed with the same accuracy when chiplet types, connection methods or placement conditions change is a separate question. The reported figure should therefore be read as a result within the scope of the conditions tested, not as an effect that applies across a wide range of product designs. When agentic AI is introduced into the design process, the key issue is not the breadth of the search but whether the constraints and models used for evaluation appropriately reflect actual manufacturing and operating conditions. The reliability of the technology as a design aid will depend on whether promising configurations were excluded from the candidates and how electrical, communications and thermal evaluations were checked and at what stage. The reported candidate reduction rate is one metric for the design process; it does not demonstrate effects on product adoption or mass production. The article is also a secondary report, and the materials available for reference here do not include an official Hitachi announcement or paper. As a result, the conditions used to calculate the candidate reduction rate, comparisons with conventional methods, measured design-time figures and details of the configurations tested cannot be confirmed. What to watch for next includes the range of design conditions covered, how evaluation results compare with actual measurements, and results from applying the technology to other chiplet configurations. Only then will it be possible to assess how more efficient candidate exploration translates into workloads and quality at design sites.

Sources

Translated from the Japanese original: /news/hitachi-chippuretto-sekkei-kouho-wo-9-wari-chou-sakugen-suru-64d3b0. The Japanese version is authoritative if they differ.