Microchip uses AKM current sensors in ML-based arc fault detection reference design

Microchip Technology has adopted Asahi Kasei Microdevices’ (AKM) CZ39 and CZ3K series coreless current sensors in a reference design that uses machine learning (ML) to detect arc faults. The design is built around Microchip’s dsPIC33A digital signal controller (DSC), which runs the signal-processing and inference algorithms on the controller itself, so detection decisions are made locally without external processing resources.

Arc faults can start electrical fires, and conventional threshold-based detection may have trouble telling a real arc from ordinary system activity. In residential and commercial AC circuits, everyday loads such as vacuum cleaners, power drills and light dimmers produce arcing at switch contacts and motor brushes that closely resembles a dangerous arc fault. In DC systems such as solar PV, EV charging and energy storage, switching transients from relay contact bounce, inverter operation and capacitor inrush generate broadband noise in the same frequency bands as an actual arc fault. Either the detector throws false positives that shut a system down unnecessarily, or its thresholds are relaxed and it misses real faults.

Microchip’s reference design runs an edge ML model directly on the dsPIC33A, using the controller’s integrated DSP engine and analog peripherals to run inference on-chip. AKM says the approach reduces false triggers and improves detection performance compared with a threshold-based design. A slow or noisy sensor blurs or buries an arc’s signature, leaving the model with nothing clean to train on.

The CZ39 and CZ3K series are coreless sensors with a 100 ns response time. Charged previously covered AKM’s proof of concept for an eFuse in 800 V EV systems.

The San Jose engineering team at AKM Semiconductor (AKMS), AKM’s US subsidiary, worked with Microchip on the current-sensing configuration during development and validation.

The arc fault demonstration is available through Microchip’s reference design program, and AKM lists solar PV, energy storage systems, EV chargers, smart ignition systems, e-Fuse designs and residential and industrial safety switches as applications.

“We’re excited to explore how this design could be adopted for data center applications as AI workloads drive the need for higher-power-density and emerging high-voltage power architectures,” said Chris Baltar, AKMS’s Vice President of Business Development.

Source: Asahi Kasei Microdevices