Chapter 1Hardware Security

§1.1 Scope

Hardware security, as the term is used in this library, covers the security properties that come from physical objects rather than from software alone: keys that never leave a device, records anchored to a device that produced them, and measurements whose outcome depends on the physical structure of a thing. Three questions organise the chapter. What can a device attest to, and to whom? How independent is the party that verifies the attestation from the party that made it? And what can an adversary do who holds the device in their hands?

The chapter states conventions and background. Results are reported in the Findings filed under it.

§1.2 Adversary conventions

Unless a finding says otherwise, the adversary is assumed to have physical access. Three classes are distinguished. A remote adversary reaches the device only over a network or a bus. A logical-local adversary runs code on the host the device is attached to, with whatever privileges that host grants. A physical adversary holds the device, can open it, and can apply instruments to it. Claims are stated against the strongest class they survive; a claim that holds only against remote adversaries is labelled as such.

Two further conventions govern how evidence is treated. First, the party that verifies a claim never holds the keys that make it. An audit layer that also holds client keys can manufacture the evidence it is supposed to check, so its verdicts carry no weight; SSX360 operates under this rule as a hard constraint. Second, a verifier is implemented independently of the writer. Where a system writes a ledger, the program that checks the ledger is written separately, ideally in a different language by a different person, so that a defect in one does not silently excuse the same defect in the other.

§1.3 Roots of trust

A root of trust is the component whose correct behaviour every other assurance depends on and which cannot itself be checked from inside the system. In practice it is a signing key and the procedure that guards it. Three arrangements recur.

A ceremony signer is an air-gapped device used under witness. Keys are generated on the device, never exported, and used only during a recorded procedure whose transcript is itself signed and published. The transcript is the evidence; the device is only its instrument.

A resident signer lives inside the system it attests for, signing each action as it happens. Its value is timeliness: the signature is made at the moment of the act rather than reconstructed afterwards. Its weakness is proximity: whatever compromises the host may reach the signer's interface, which is why per-actor sequence numbers and gap detection (§2.3) matter more for resident signers than for ceremony signers.

A hardware-anchored record is a ledger entry that includes an attestation from one of the above, so that a reader can tell not only who signed but on what. In the evidence levels of §2.6, hardware anchoring is the highest of the three.

§1.4 Measurement as evidence

A physical measurement can serve as evidence about a device in the same way a signature serves as evidence about a message, provided it satisfies the same discipline. The measurement must be reproducible: repeated by the same operator on the same instrument, it returns the same result within a stated tolerance. It must be calibrated: the instrument is referenced to a standard before and after, and the reference is recorded. It must be independently repeatable: a second laboratory, given the sample and the protocol and nothing else, obtains the same result. And it must be recorded as it is made: each acquisition is written as a signed receipt to an append-only ledger with the instrument, settings, sample identifier, and operator, so that the ordering of measurements and any omissions are later checkable.

Raman spectroscopy of carbon thin films is the measurement this library is most concerned with; Chapter 3 describes the physics and the pipeline. The general point is independent of the instrument: a measurement that is not reproducible, calibrated, independently repeated, and recorded at the moment of acquisition is an observation, not evidence.

§1.5 Evidence properties

Evidence about an action or a device is classified here by three properties, each of which a third party can check mechanically.

Signed. The record carries a signature from a key whose holder is identified. A verifier can establish who vouched for the record and that it has not changed since.

Gap-checkable. Records from a given actor carry contiguous sequence numbers, so that a verifier can tell whether any record has been withheld. A signed record proves what was said; a gap-checkable series proves that nothing was left out.

Hardware-anchored. The signing key is held in hardware of a known kind under a stated custody procedure, and the record includes an attestation to that effect.

The properties are cumulative: a hardware-anchored record is also signed and gap-checkable. They are the basis of the MCV-1 criteria described in §2.6.

§1.6 Reading

Anderson's Security Engineering [A20] remains the broadest treatment of the subject and of the ways in which physical and procedural controls fail. The cryptographic primitives assumed throughout, SHA-256 and Ed25519, are specified in [RFC6234] and [RFC8032]; the design rationale for Ed25519 is given in [BDLSY12].