Skip to content

What is a security chiplet, four functions in one boundary

A security chiplet places entropy generation, runtime verification, root of trust with key storage, and cryptographic execution inside one boundary, as a die that can be integrated into a larger package. It turns a set of components a customer assembles into a block they place.

Four functions, one boundary

  • Entropy generation — a physical source, on the die, available at first boot.
  • Runtime verification — measurement of the entropy source's own health, so the entropy claim is current rather than historical.
  • Root of trust and key storage — the anchor and the keys it protects.
  • Cryptographic execution — the operations those keys are used for.

Assembled as separate components, every interface between these is a boundary that must be crossed, protected and evaluated. Co-designed into one die, the interfaces are internal.

Why chiplets rather than a discrete part

Chiplet integration has become normal in high-performance silicon: a package holds several dies, each on the process that suits it, connected by a standardised interconnect. That changes what a security function can be.

It stops being a component on a board with a bus between it and the processor, and becomes a die in the same package — shorter interfaces, less exposed, and integrated by the customer as a block rather than as a subsystem.

What co-design buys that assembly does not

Mainly, that the entropy source and the thing verifying it are designed together. Runtime verification of an entropy source needs access to physical measurements from inside it, which is straightforward on one die and awkward across a package boundary.

It also means key material never crosses an external interface between generation and use, which removes a set of questions an evaluator would otherwise ask.

Where the entropy source has to be

Inside. An entropy source outside the boundary means entropy crossing into it, which is the trust-boundary problem again in miniature — and the same argument applies at chip scale.

This is viable because the source became small enough to integrate. The engineering behind that is what makes the chiplet argument possible rather than aspirational, and IP licensing is the other route to the same place.