High Performance Physical AI DevKits.

Ultra-low-power (<10W) silicon, a carrier board with rich I/O, and pre-installed Palette™ software — everything you need to build and deploy physical AI quickly.

Buy DevKit (US & Europe) Request DevKit (Other Regions)

MLSoC 50 TOPS

Perf / Watt Industry leading

Models CNNs to LMMs

Power Less than 10W

Build and deploy your applications seamlessly with industry leading performance per watt

The SiMa.ai DevKit leverages the purpose-built MLSoC and Palette Software™ that includes our agentic development environment, Palette Neat. Compile and execute complex, multimodal AI with minimal effort and out-of-the-box high performance. Learn more at our Developer Center.

SiMa.ai Palette Neat

Explorer

▾ Compile Models

PY model_compile.py

▾ Run Python App

PY app.py

▾ Run c++ app

CMAKE CMakeLists.txt

CPP main.cpp

model_compile.py

1 # Compile your First ONNX Model

2

3 # 1. Load the ONNX model.

4 importer = onnx_source(

5 args.model,

6 {args.input_name: INPUT_SHAPE},

7 {args.input_name: ScalarType.float32},

8 )

9 loaded_net = load_model(importer, target=target)

10 log.info("Loaded %s for %s", args.model, args.device)

11

12 # 2. Prepare the calibration dataset.

13 calib_images = load_calibration_images(args.calib_images,

14 calib_data = convert_data_generator_to_iterable(

15 DataGenerator({args.input_name: calib_images}))

Supports 200+ ONNX-compatible / Hugging Face models, including CNN models (ResNet, YOLO, SegFormer, etc.) and GenAI models (Gemma 4, Qwen 3.5-VL, etc.).
Mixed-precision quantization with < 1% accuracy delta.
Outputs a hardware-optimized .sima binary — no manual tuning required.