Embedded AI on ESP32 Board

Real-time BGA void detection on a microcontroller

Sub-milliscecond inference. No cloud. No GPU. Just a microcontoller.

The opening shot is a high-quality CT scan of the ESP32-S3-Nano microcontroller, i.e.,  the exact board running the inference in this demo. The board detects voids in BGA solder joints from X-ray data in under 1 ms, running a trained autoencoder entirely on-chip at 240 MHz. 

Key Facts

< 1 ms

AUC 0.81

Open Source

Inference time on

ESP32-S3 @ 240 MHz

Calibrated for voids exceeding 10 % threshold

Full source on GitHub


Project Description

X-Inspector

The project repository contains an embedded AI demo for detecting voids in Ball Grid Array (BGA) solder joints. It demonstrates high-speed quality inspection using an ESP32-S3 to analyze X-ray data at the edge.

This project is a streamlined version of some of the core inspection principles used in the industrial X-ray analytics platform of maXerial.

Try It Yourself

The full firmware and Python desktop app are available on GitHub github.com/maxerial/embedded-ml-on-esp32-demo

This project is a research and community demo by maXerial AG, Vaduz, Liechtenstein. For industrial-grade BGA inspecation, see Falknis X-ray Inspect.