Eta Compute

Let’s rewind two years. Back then, developing a demo or a customer application at Eta Compute was an iterative and experimental process - essentially a cycle of trial and error: - Choosing a model architecture and testing iteratively to see if it met target and application requirements. (Does it run on chip X? Is it fast enough for the application?) - Training the model, which involved developing or adjusting training code and provisioning hardware to speed up experiments. - Evaluating model accuracy—and if the results weren’t good enough, going back to step 1. This loop continued until we achieved a satisfactory outcome. This process consumed around 70-80% of the team’s effort, with the remaining time split between firmware, application code, UI, testing, and other tasks. Then, at Eta Compute, we developed Aptos, our no-code Edge AI platform, to accelerate neural network development for embedded systems. Today, the effort required for model development in this particular demo has dropped to just 5%. It’s as simple as: - Uploading your data to the Aptos platform - Selecting your target chip (in this case, the Silicon Labs MG26) - Defining constraints (memory limit, runtime, input resolution, etc.) Pressing the "GO" button Eight hours later, I had 50 models to choose from—each fully characterized on the MG26 chip and with guaranteed accuracy. 50 models! With the right tools, Edge AI has never been easier! The team can now focus on real application development instead of getting lost in the complexities of data science.

It was a terrific week at Embedded World 2025!??Here is a brief video with our CEO Evan D. Petridis showing one of our demos.?A big thank you to our partner Silicon Labs for the collaboration!?Please contact us to see the full demo + see how we created the multiple models quickly using Aptos ML tools.

If you are attending EDGE AI FOUNDATION Austin 2025 this week, be sure to come by and see our Vision AI demonstration, with low power ML models built by Aptos specifically for the Silicon Labs EFR32xG26! We're showing the demonstration at the Silicon Labs table in the exhibitor area. Evan D. Petridis Tamas Daranyi

We're proud to announce our new partnership with Silicon Labs! Read more about it below. Or visit our website for a free trial of Aptos and you can generate your own optimized ML models targeting Silicon Labs solutions.

The future of ML is Tiny! I have the pleasure to participate in tinyML Foundation EMEA 2024 Summit in Milan on behalf of Eta Compute. We're exhibiting a computer vision demo together with Arrow Electronics, developed using our no-code Edge AI development tool: Aptos. It recognizes hand gestures at an astounding rate of 30 fps on a NXP Semiconductors' tiny MCX-N NPU (Neural Processing Unit).

Announcing Eta Compute’s Collaboration with Qualcomm for Edge ML! ?? Developers rejoice! ?? Eta Compute’s #Aptos software now supports Qualcomm Technologies’ IoT products, offering a streamlined path to build high-performance machine learning (ML) models running on Qualcomm Technologies silicon. This collaboration between the companies aims to accelerate machine learning product development at the Connected Edge! ??? Eta Compute’s focus is on closing the gap between ML and embedded systems with our no-code edge ML toolchain, Aptos.?Aptos shrinks edge ML development time from months to days, while generating faster, higher-accuracy models.?Qualcomm Technologies’ IoT solutions enable the real-world Connected Intelligent Edge, accelerating digital transformation across many industries. “Eta Compute’s Aptos learns the capabilities of devices powered by our technology and applies this knowledge to the automated creation of high-performance ML models,” said Manvinder (Manny) Singh, Vice President, Strategy and Partner Solutions Development, IoT at Qualcomm Technologies, Inc. “We look forward to working together to assist our customers who are creating the exciting AI-powered products of the future on the Connected Edge.” “Qualcomm Technologies is a leader in providing powerful heterogeneous computing platforms for the Connected Edge,” said Evan D. Petridis, CEO of Eta Compute.?“Aptos enables users to take full advantage of Qualcomm Technologies’ silicon by automatically and quickly generating optimal edge ML models to run on any of the compute cores on devices powered by Qualcomm Technologies – or quickly try them all and evaluate which is best suited for their application.” Eta Compute's Aptos platform now supports the Qualcomm? AI Stack, a unified AI portfolio that works across multiple chipsets across a broad range of capabilities. Uniquely, Eta Compute’s Aptos allows developers to target the ML model to any or all of the processing engines in a Qualcomm Technologies powered device, including CPU, GPU, DSP and AIP cores.?This dramatically simplifies the system design task, while improving real-world performance. Eta Compute, in collaboration with Qualcomm Technologies, has defined a roadmap for expanded support of Qualcomm Technologies’ silicon during 2024. In one example of the collaboration, Eta Compute onboarded the Qualcomm? QRB5165 Processor onto Aptos and made the integration available to mutual customers.?The QRB5165 Processor brings the latest in IoT technologies into a highly integrated chipset to power the next generation of high compute, AI-enabled, low power robots and drones for consumer, enterprise, defense, industrial and professional services sectors that can be connected by 5G. ??Together, we're redefining the future of Edge ML! ?? #EdgeAI #Innovation #Qualcomm #ETACompute

At the embedded world Exhibition&Conference in Nuremberg this week, our team had a hit with a fun hand recognition slot machine game on the show floor! The AI model for the game was developed entirely (and quickly in 18 hours) using our own #Aptos Edge AI development platform, trained with pictures of hand gestures and directed to target the NXP Semiconductors MCX N947 and eIQ? Neutron NPU.