Skip to content

IN DEVELOPMENT · NO CHIP HAS BEEN TAPED OUT

A machine that can prove what it is

Before a site lets a machine on its network, it wants to know what the machine is and what it is running. The first E31 design includes a hardware root of trust so that a machine can answer.

· 4 min read

IN SHORT

Identity is one of the four jobs BramHA is being designed to do. The first design includes a hardware root of trust, intended to let a machine prove what it is and what it is running before a site lets it on the network. The decision to admit the machine stays with the site. No chip exists, and no security certification is claimed.

A facility's network is where its machines meet. Machines from different vendors join it, and through it they reach the inference pool in the building. Before a site lets a machine on, it has a fair question: what is this machine, and what is it running?

Identity is one of the four jobs BramHA is being designed to do. The program is in development and no chip exists. This article describes what a hardware root of trust is and what the program intends it to do.

The question a site asks

A machine that joins a network usually identifies itself by something it has been given: an address, a name, a credential stored in its software. Each of those can be copied. A site that accepts them is trusting that nothing has been copied or changed.

For a laptop, that may be an acceptable risk. For a machine that moves on a floor where people work, a site may want something firmer. It may want to know that the machine is the one its vendor built, and that it is running the software its vendor released.

Those are two separate questions. The first is about the hardware. The second is about what the hardware has loaded.

What a hardware root of trust is

A root of trust is the part of a system that everything else is checked against. It is called a root because nothing beneath it needs to be trusted first.

A hardware root of trust places that part in the chip. In general terms, it holds a secret that software cannot read out or change, and it is the first thing to act when the chip starts. Because it sits in hardware, replacing the machine's software does not replace it.

The glossary entry on root of trust gives the short definition.

What it is, and what it is running

The program's intent is that the root of trust lets a machine answer both of the site's questions.

What it is: the machine can respond to a challenge in a way that only a part holding its secret could. A copy of its name or address is not enough.

What it is running: the chip can check the software it starts, and can report what it started in a form that the software itself cannot forge.

A site can then check those answers before it lets the machine on the network. The check is the site's, and so is the decision. The silicon is intended to give the site something to check. It does not decide who is admitted.

This matters most in a facility with machines from many vendors, which is the kind of facility a Czar Grid is built to serve. How a site's own systems would use the proof has not been published.

Why it is in the first design

The first design is deliberately small: a real-time controller with a root of trust. It may seem odd to pair the two in a design that leaves so much else out.

The reason is that identity is hard to add afterward. A root of trust has to be present from the first moment a chip starts, or there is a gap before it in which anything could have run. It belongs in the foundation of a design, not in a later revision.

It is also a job that suits the program's choices. Proving identity needs care in design and verification. It does not need dense computing or the leading edge of chip manufacturing.

What identity does not do

Identity is not safety. A machine that proves what it is can still be badly programmed, and a machine that proves what it runs can still run software with faults. Safety, stops, and motion stay on the machine as its vendor built them, and the safety case is the vendor's. Where safety lives covers that boundary.

It is not a replacement for a site's network security. It is one input to it.

And it is not certified. No chip has been taped out, the specification of the first design is not published, and no security evaluation or safety certification is claimed. The page at /chip describes the four jobs, and nothing there can be ordered.

Questions this raises

Why does identity need hardware?
Software can be copied or replaced, so a claim made only in software can be made by anything. A root of trust in hardware holds a secret that software cannot read out or change, which gives the machine something only it can show.
Does a root of trust make a machine safe?
No. It is intended to show what a machine is and what it runs. Whether that machine behaves safely is a matter of its vendor's design, and the safety case stays the vendor's.
Has the root of trust been evaluated or certified?
No. No chip has been taped out, the specification of the first design is not published, and no security or safety certification is claimed.

KEEP READING