From trained model to running hardware
ECS-Studio is the development suite for ECS-DoT. It brings the whole edge AI workflow into one workspace: port a trained model with ECS-Compiler, measure it with ECS-Profiler, start from ECS-Model Zoo, then build, flash and debug on real hardware. It installs in one step on Windows, macOS and Linux, with no manual toolchain setup and no command-line work.
THE WORKSPACE
A VS Code-based IDE with the ECS-DoT SDK, toolchain and sensor libraries built in. One-click build, flash and debug, plus a serial monitor and live plotter.
MODEL PORTING
Imports TensorFlow Lite, PyTorch and ONNX models, quantizes them to INT8, maps each layer to the deep-learning accelerators or the RISC-V core, and generates a ready-to-use inference library.
PERFORMANCE INSIGHT
Profiles your model running on ECS-DoT hardware and visualizes the results, so you can see latency and memory use and compare model versions before you commit.
PRE-PORTED MODELS
Audio, vision and time-series models already ported to ECS-DoT, available to registered users in the ECS-Studio portal — run them as-is or use them as the starting point for your own.
Import models trained in TensorFlow Lite, PyTorch or ONNX. ECS-Compiler quantizes them to INT8 and maps them onto ECS-DoT.
A pre-check shows where every layer will run — AI accelerator or RISC-V core — and whether the model fits in on-chip memory.
Operations the accelerators don’t handle run on the RISC-V core in the same firmware image, so more models deploy as-is.
Your model lands in the project as a library with a single inference call, an example application and a latency benchmark.
Build, upload, monitor and debug from the toolbar, with full hardware debugging on ECS-DoT silicon and a live serial plotter.
Native installers for Windows, macOS and Linux. The Toolchain Manager installs and verifies the compiler, SDK and debug tools.
From trained model to running inference on ECS-DoT
Checks the model layer by layer: what runs on the accelerator, what runs on the RISC-V core, and whether it fits in memory
Add an optional calibration dataset of real data from your use case
8-bit (INT8) quantization, with min-max, entropy, percentile or automatic calibration
Converts the model to an optimized intermediate representation
Generates the C code that runs the network on ECS-DoT
A library, header, example app and latency benchmark, copied straight into your project
See how your model really runs
Run your model on ECS-DoT hardware and see where the time and memory go. ECS-Profiler measures inference latency and memory use on target and visualizes the results, so you can compare model versions and tune before you commit to one.
Start from a model that already runs on ECS-DoT
A growing library of models ported to and validated on ECS-DoT, across audio, vision and time-series — keyword spotting among them. Available to registered users in the ECS-Studio portal: open a model in ECS-Studio and run it on your board, or use it as the starting point for your own.
Registered users also get ECS-Model Zoo — pre-ported audio, vision and time-series models ready to run — or bring your own model with ECS-Compiler.