Forging Intelligence In circuits woven, thought takes flight, A symphony of logic, fast and light. Systolic pulses, data streams, Shaping visions once seen in dreams. Bridges swift, precision wide, Scaling knowledge, side by side. Transformer sparks, the code unfolds, A dance of power in patterns bold. Boundless minds in silicon cast, Tomorrow's genius, built to last.

Open Source Toolkit

The SiG Development Kit provides a robust solution for model porting and optimization on the powerful and efficient BPU architecture.

Key Features:
Seamless Framework Support – Compatible with PyTorch, TensorFlow, and ONNX, enabling efficient model training, quantization, optimization, and deployment.

Automatic Optimizing Compiler – Ensures maximum performance on BPU hardware accelerators with minimal developer intervention.

High-Efficiency AI Processing – Maintains accuracy and computational efficiency, optimizing deep learning workloads for SiG computing solutions.

Minimum Effort. Maximum Performance

Typical porting time​
0 weeks
Successful customer projects
0 +
Productivity improvement
0 x
Loss of accuracy
Near 0

Train, Deploy & Optimize

Train & Quantize

Two quantization methods are supported:

  • Post-Training Quantization (PTQ) – Enables rapid conversion from floating-point to fixed-point representation, allowing quick trials and evaluations.

  • Quantization-Aware Training (QAT) – Optimizes models during training to minimize accuracy loss, ensuring near-zero degradation in precision.

These techniques enhance efficiency while maintaining model performance.

The model compiler automatically optimizes execution by applying techniques such as kernel and feature map splitting, vertical and horizontal layer fusion, and instruction grouping and scheduling. These enhancements:

  • Increase parallelism – Boost data and instruction efficiency for faster processing.

  • Reduce external memory access – Minimizing memory visits lowers latency and power consumption.

  • Enhance computational efficiency – Streamlining execution ensures optimal performance on hardware accelerators.

The SDK provides comprehensive documentation, reference sample applications, and pre-implemented algorithms tailored for automotive use cases. It accelerates software deployment by offering:

  • Extensive Documentation – Detailed guides for seamless integration.

  • Reference Samples – Ready-to-use applications for rapid development.

  • Algorithm Implementations – Optimized solutions for automotive AI and automation.

TogetheROS

A Set of Safe, Simple and Easy-to-use Smart Driving
Embedded Middleware

TogetheROS Auto Application Development Suite

The TROS.A Automotive Development Suite provides a safe, reliable, open, and user-friendly software environment for efficient development, integration, and testing on Journey solutions.

Key Features:

  • Comprehensive Framework & Services – Supports seamless application development with built-in tools.

  • Example Applications & Verification Toolset – Accelerates development with pre-built samples and robust testing capabilities.

  • Standardized Interfaces & Protocols – Ensures compatibility with third-party software components for smooth integration.

"Enhancing Safety and Quality of Life with Smart Driving Vehicles."​

"Exploring Our Journey Together"