Technology

Purpose-built silicon for cryptography's hardest math

General-purpose processors hit a wall on cryptographic workloads. We design the circuits that don't.

Why custom hardware

Cryptographic workloads keep growing — signature verification, zero-knowledge proofs. CPUs run out of throughput; GPUs improve it but carry the overhead of generality: power-hungry, physically large, and not shaped like the math.

Purpose-built silicon eliminates that overhead, and the advantage scales in both directions — ultra-low-power operation at the edge, or maximum throughput in the data center.

This already happened once

Cryptocurrency has already run the experiment. Bitcoin mining is one cryptographic operation — SHA-256 — with an open market incentive attached, and fifteen years of competition moved it from Satoshi's CPU through GPUs to purpose-built ASICs:

SHA-256 mining efficiency — hashes per joule
100 K
1 M
10 M
100 M
1 B
10 B
100 B 10 K
20 B
40 B
60 B 0
CPU · 2009first GPUsfirst ASICsASICs · 2024
Logarithmic scale — each labeled gridline is ×10 the one below. Toggle to linear to see the same data at true proportions.
SHA-256 mining efficiency by hardware generation, hashes per joule
HardwareHashes per jouleVersus 2009 CPU
CPU · 2009100 Kthe starting point
first GPUs1.2 M12× the 2009 CPU
first ASICs500 M5,000× the 2009 CPU
ASICs · 202460 B600,000× the 2009 CPU

Fig. 5 A ∼600,000× efficiency improvement in fifteen years. Our founder's 2024 Sydney talk walks through the numbers, drawn from the public mining-hardware record.

Over those same fifteen years, the network's transaction capacity did not move: roughly seven transactions per second then, roughly seven now. Had validation improved with mining, Bitcoin would process millions of transactions per second today. The force that produced a 600,000× improvement was real — it was simply pointed at a hash puzzle that exercises none of the machinery that validates transactions, so node capacity stayed decoupled from it.

We take the same trajectory — purpose-built silicon, relentlessly optimized for a single operation — and point it at the math the network actually depends on: signature verification.

The engine that does it is the K1; the open platform it drops into — and what we open-source — is on the Open Source page.