Hardware or software root of trust
A software root of trust relies on the platform's existing protections to keep its key secret; a hardware one puts the key somewhere software cannot reach. Software is a defensible choice when the threat model excludes physical access and the platform already isolates well, and it stops being one when either assumption fails.
What a software root of trust is
Verification logic and a key held in software, protected by whatever the platform offers — a hypervisor boundary, a trusted execution environment, an operating-system keystore. The chain works the same way; the anchor is enforced by other software.
This is not a straw man. Plenty of production systems are built this way and are appropriately secured for their threat model.
When it holds up
- The device is physically controlled — a machine in a datacentre you operate rather than a meter on a wall.
- The platform provides genuine isolation, and you are prepared to trust that boundary.
- The consequence of compromise is bounded and recoverable: a key you can rotate rather than an identity fused for the life of the device.
- The economics rule out per-unit hardware cost, and the alternative is no root of trust at all.
When it stops holding
When an attacker can touch the device. Physical access defeats software isolation given time, and for a device deployed in the field that time is unbounded.
It also fails when the isolation boundary you depend on is itself the thing under attack — a hypervisor escape or a TEE vulnerability takes the root of trust with it, because it was inside.
The middle ground most systems occupy
Not a discrete HSM, but a hardware security block inside the main SoC: key storage, verification logic and often an entropy source, in silicon, on the die that is already there. Cheaper than a separate component and materially stronger than software alone.
That is where most volume device design has settled, and it is why the quality of the entropy source in that block matters so much — and why it is worth asking what your silicon vendor documents about it.
The question to actually ask
Not which is better, but: if this device is in an attacker's hands for a month, what must still be true? Whatever is on that list has to be below the software boundary.
Keep reading
- Foundations
Device identity and secure provisioning
A device identity is a key pair generated inside the device and bound to it, which the device uses to prove what it is. Its strength is fixed at the moment of generation: entropy available at first provisioning determines whether the identity is unique and unpredictable for the device's entire life.
- Foundations
What is a hardware root of trust, and what it actually protects
A hardware root of trust is an immutable anchor — typically a key and verification logic in silicon — that the rest of a system's trust chain is built on. It establishes what code was allowed to run and protects the key material underneath, and it cannot verify anything about its own correctness.
- Foundations
Hardware security or software security, framing the question properly
Hardware and software security are layers, not rivals. Cryptography always executes on physical hardware, so the question is never which to use but which properties must be enforced physically — key isolation, tamper response and unpredictable key generation — and which can be enforced in code.