In-house silicon

Notes

The wave of hyperscaler and handset in-house accelerators from the 2010s onwards reshaped the customer base of the merchant chip industry.

Overview

In-house silicon means chips that a systems, cloud, consumer or automotive company designs for its own products instead of buying standard parts from a chip vendor. The company owns the architecture and design but usually relies on foundries, assembly and test partners and licensed IP to build them. It is sometimes called custom or bespoke silicon, and companies pursue it for differentiation, cost and power at scale, and closer control of hardware and software together.

How it works

A company starts down this path when off-the-shelf chips do not match what its products need, or when its volumes are large enough to justify the design cost. It builds a chip team, licenses processor cores and interface IP, uses commercial EDA tools, and contracts a foundry for wafers and an OSAT or the foundry for packaging. Some companies instead work with an ASIC design services partner that handles parts of the implementation.

The economics differ from those of a chip vendor. A merchant chip must be sold to many customers to recover its development cost, whereas an in-house chip is paid for by the owner's own products: cost can be spread across thousands of servers or millions of devices, and the design need not appeal to anyone else. That allows highly specific features, tight integration with the company's software and secrecy about how it works. Supply control is another motive: a design made exclusively for one company gives it a captive supply route.

The approach has limits. It needs scarce design talent and enough volume, and vertical integration does not suit every company or every hardware need, so many keep buying merchant chips for some functions. Licensed IP and design services help newcomers, and demand for them has grown with this trend.

Examples

• Apple's M1, announced in November 2020 as the first chip Apple designed specifically for the Mac, beginning the Mac's move to Apple silicon.
• Google's Tensor Processing Units, custom-developed ASICs used to accelerate machine learning in Google Cloud.
• Hyperscalers such as Amazon and Meta designing chips for their own data centres, as described by Semiconductor Engineering.
• An automaker developing its own chips so that hardware is tightly integrated with its software-defined vehicle platform.

Why it matters

In-house silicon has shifted part of the chip industry's design work from chip companies to their largest customers. Custom silicon architects, programme managers, sourcing leads and design services teams are in demand, and chip vendors increasingly compete with, or sell IP and services to, their own customers.

Related terms

• Custom silicon
• Merchant silicon
• Vertical integration
• Hardware-software co-design

Where it appears on the Hub

Used on 6 pages of the Hub. In each group, the pages that mention it most come first.