AI Toolchains

Ashling, in partnership with Embecosm, delivers advanced AI tooling infrastructure and services that seamlessly bridge model development and device execution. Our expertise spans graph optimization, memory planning, and performance tuning, with broad support for PyTorch export, TensorFlow, SYCL, OpenCL, IREE, OpenXLA, LLVM-based compilation, and ExecuTorch on microcontrollers. Together, we enable dependable, scalable, and production-ready toolchains that accelerate innovation from lab to deployment.

We’re currently working on a number of AI related services projects for our partners including the following:  

  • PyTorch and ExecuTorch support for smaller, edge-based, microcontroller platforms capable of running neural network AI models which are particularly suitable for audio and video analysis at lower resolution. ExecuTorch runs “bare-metal” on the microcontroller so no OS (e.g. Android or Linux) required.
  • Extending the RISC-V GCC toolchain with support for custom ISA extensions allowing targeting and control of AMD GPUs via the ROCm software stack for AI.
  • Multi-core, parallel debugging support for an AI Inference accelerator using over 500+ custom dual-core, multi-threaded processors designed to achieve high accuracy in convolutional neural networks. Debug is via a direct PCIe link using a Visual Studio Code Extension based UI with a highly intuitive and innovative graphical and command-line based interface.
  • Custom SDK for a partner’s new SoC containing both Arm’s Cortex M85 Microcontroller and the Ethos U65 Neural Processing Unit (NPU).
Cortex-M85 MCU with Ethos U65 NPU based SoC
Simultaneous Debugging of Hundreds of Cores in an AI Accelerator
AMD GPU control via ROCm AI Software Stack

A Proven Partnership

Embecosm’s expertise in compiler and AI toolchains perfectly complements Ashling’s embedded development tools and services, including IDEs, debuggers, SDKs, debug and trace probes, simulators (e.g. QEMU & Verilator) and development kits. Over more than eight years of partnership, Ashling has represented Embecosm on multiple toolchain projects, many focused-on RISC-V architectures. Together we operate globally from offices in the US, India, UK, Ireland, France, and Germany, serving customers from Tier-1 semiconductor companies to start-ups in stealth mode. Joint customers to date include companies such as: Akeana, Altera (Intel), Cadence, OpenHW Group, RISE, Rivos, Tenstorrent and Western Digital.  

Book an engineer-to-engineer call with Ashling and Embecosm to learn how we can help with your AI toolchain and project needs. Just email: engineer@ashling.com