Drag to turn the package
Onyx A1
In Production
The flagship. 16 dedicated NPU cores run generative inference on-device, no fan, no cloud round-trip.
- Process node
- 3nm · Frontier
- Category
- Edge AI SoC
- Cores
- 8 CPU + 16-core NPU
- Package
- FCBGA-620
- Power
- 6.5W typical
- Die size
- 88 mm²
Datasheets are shared under NDA. Applications engineering typically responds within one business day.
About this part
Onyx is the SoC we built for phones and tablets that need to run real generative models locally. The GAA transistors at 3nm buy us the density for a 16-core NPU block alongside 8 general-purpose cores, all inside a 6.5W typical envelope — the kind of budget a passively-cooled device can actually sustain.
Most of the design effort went into the NPU’s memory fabric, not the compute cores themselves: keeping activations resident on-die is what actually determines whether a model runs at a usable token rate or thermal-throttles after ninety seconds. Onyx ships with a full toolchain for quantizing and mapping common model architectures onto that fabric directly.
In production since Q2, currently in three shipping consumer devices with two more designed in.