SANTA CLARA, California—Startup Eta Compute has pivoted, once again, to offer a hardware-aware AI model optimization tool for tinyML applications and has onboarded NXP’s MCX N microcontroller with NXP’s Neutron AI accelerator onto its platform. Eta Compute’s platform, Aptos, is a no-code MLOps toolchain that simplifies model development by optimizing model architecture for the particular embedded hardware target.

“The MCX N chips are industry-leading in their category, and we are already seeing pretty great performance out of Aptos for them,” Eta Compute CEO Evan Petridis told EE Times.

Eta Compute started as an ultra-low–power AI chip company. The Eta Compute design was a dual-core Arm CPU plus NXP DSP design, and part of the company’s secret sauce was the software, which divided the workload between the cores. In 2020, Synaptics invested in Eta Compute and took an exclusive license for Eta Compute’s Tensai Flow neural network compiler, which Synaptics now supplies with its Katana AI-enabled SoC line. Petridis said Eta Compute is no longer working on Tensai Flow, as that has been handed over to Synaptics. Skip ahead to 2023, and Eta Compute has pivoted again—this time emerging with an MLOps toolchain for embedded systems development, Aptos.

NXP’s MCX N microcontrollers with neural processing unit join Synaptics’ Katana SoC as the first hardware targets onboarded onto Eta Compute’s Aptos toolchain to date.

 ----Form EE Times