Image: reddit.com · Video still: ML-KEM-1024 on an ESP32: 34-40 ms per key exchange, and the stack is what runs out first
Post-quantum ML-KEM-1024 key exchange measured on ESP32 boards
A master's thesis project running ML-KEM-1024 post-quantum key exchange on ESP32-S3 and C6 boards, with about 60% of the code written with Claude Opus 5.
Credits
- Reddit poster Relative-Ad-9876 Source ↗
Earliest dated source:
The full story
For a master's thesis the maker ran post-quantum key exchange (ML-KEM-1024) on XIAO ESP32-C6 and ESP32-S3 boards, as part of a full protocol against a Go server over Wi-Fi. Measured on real hardware, encapsulation took 18.4 ms on the S3 and 16.0 ms on the C6, decapsulation 21.1 ms and 17.8 ms, and the post reports 34 to 40 ms per key exchange, with stack space running out first.
They say roughly 60% of the code was written with Claude Opus 5's involvement, and the post itself was also written with Claude's help. The problem framing, algorithm and architecture choices, test setup and measurements were theirs.
How it was done
Tools
Themes
Loading themes…
Sources
From a short post by the maker, so the details are as they tell them.
- Reddit: ML-KEM-1024 on an ESP32: 34-40 ms per key exchange, and the stack is what runs out firstReddit poster: Relative-Ad-9876Posted
The maker's post on r/esp32.
If it disappears: archived copy of Reddit: ML-KEM-1024 on an ESP32: 34-40 ms per key exchange, and the stack is what runs out first (opens in a new tab)
- Reddit: Post-quantum crypto on ESP32: what ML-KEM-1024 actually costs on a C6 and an S3
The poster has deleted it, but the original is still readable (it can take a few seconds to load): archived copy of Reddit: Post-quantum crypto on ESP32: what ML-KEM-1024 actually costs on a C6 and an S3 (opens in a new tab)