Skip to content

From crypto agility to hardware agility

The industry accepted years ago that software must be able to change its cryptography. The same argument applies with more force to hardware, which changes on a decade-long refresh cycle rather than a release cycle — and where the inability to change is decided at design time by someone else.

Crypto agility is an accepted idea. Everyone nods at it, most architectures aspire to it, and the post-quantum transition has turned it from good practice into an operational requirement.

The version of the idea in circulation stops at the software boundary, and that is where it stops being useful.

The argument that is already accepted

Cryptographic algorithms have finite lives. Assumptions weaken, cryptanalysis advances, standards are revised. A system that hardcodes an algorithm will eventually be a system that must be redesigned, at a moment chosen by events rather than by you.

Nobody disputes this. It is why algorithm identifiers are negotiated, why libraries are abstracted behind interfaces, and why "we can swap the cipher" is a design goal.

The extension nobody makes

Every one of those arguments applies to hardware, and each applies more strongly.

Software changes on a release cycle measured in weeks. Hardware changes on a refresh cycle measured in years, and in industrial, automotive and infrastructure contexts, decades. If the argument for agility is that the world changes faster than your system does, the gap is far wider for hardware.

Software agility is also within your control — expensive refactoring, but yours. Hardware agility was decided years ago, by a designer at a vendor, and no amount of budget retrofits it. You inherit the decision and discover it during a migration.

What it means concretely

That cryptographic functions in hardware should be implemented in something that can be reprogrammed after deployment rather than committed to fixed silicon. When a parameter set is revised, that is a firmware update to a device already qualified and already installed, rather than a respin, a requalification and a customer upgrade programme — the cost difference is not incremental.

The honest counter-argument

Programmability costs area, power and unit price. For a high-volume, cost-dominated, power-constrained device implementing a stable algorithm, fixed silicon is the right engineering answer and pretending otherwise would be selling.

The trade turns on deployment life and change probability. Short-lived and stable favours fixed. Long-lived, in a regulated estate, during a standards transition on assumptions younger than the devices — that favours programmable, and by a margin that is not close.

Where this lands

Crypto agility, taken seriously, is a procurement requirement rather than an architecture principle. It means asking a hardware vendor not only which algorithms they support but whether support can change after you buy, and what that change costs you in requalification.

Very few buyers ask that today. The ones migrating an appliance estate before 2030 will learn to, because they are about to find out which of their devices can move and which cannot — and the inventory is where they find out.