Chiplets Ease Pain for Large Designs

Chiplet use is increasing, primarily for large in-house computing projects, but small chip designs don’t benefit. A broad third-party chiplet market remains years away.

Bryon Moyer

Chiplets are receiving greater attention now that many leading semiconductor companies have begun to employ them. Although barriers to adoption remain, they’re dropping even as the appropriate role for chiplets comes into greater focus. Leading-edge processors with large transistor counts will increasingly benefit from this technology, although most will be single-vendor designs.

Chiplets have been captive to large companies that disaggregate big computing chips into smaller blocks that can accomplish the same tasks more economically. The technology has gained urgency as the 3nm node delivers fewer benefits than expected. Each vendor has a twist on how best to combine chiplets in advanced packages; together, they’re charting a path toward something that could become a common process.

Ongoing challenges, discussed at the recent Linley Fall Processor Conference, indicate chiplet proliferation will be slower than some advocates project; a keynote by Cisco’s Ranganathan Sudhakar summarized elements of both hype and hope. His comments reflected Cisco’s experience with several chiplet generations as far back as 2010. Cost remains a primary concern, and chiplet benefits accrue almost exclusively to large designs such as high-speed processors. They’re unlikely to make sense as a vehicle for selling small hardened intellectual-property (IP) blocks.

A multivendor chiplet marketplace, often touted as a long-term goal, appears unlikely for several years. A universal approach to chiplet design and assembly seems unfeasible, although the industry could endorse a set of standards that address a wide range of applications in an orderly way.