ztachip Documentation

A multicore, data-aware, embedded RISC-V accelerator for edge vision and AI inference. Fully open source — hardware and software — under the Apache 2.0 license.

Getting Started: Running a Vision/AI Demo on Artix-7 FPGA →Everything needed to see ztachip working on real hardware: building the software and the RISC-V toolchain, building the FPGA image, wiring up the camera and VGA modules on a Digilent Arty A7 board, and loading and running the vision and LLM demos, with the benchmark results measured on that board. The Arty A7 is only the reference platform — the guide also covers porting ztachip to another FPGA, ASIC or SoC, and running it in simulation with no hardware at all. Technical Overview →What ztachip is, the problem it solves and what it achieves. Explains why a domain-specific architecture is needed for edge AI, what the project provides, and how it compares on performance and power. Hardware Architecture →How the hardware is put together, component by component: the tensor processor that coordinates everything, the engines that move tensor data to and from memory, the floating point unit, and the array of VLIW cores that do the arithmetic. Each component links to its RTL source. Programmer's Guide →How to program ztachip directly, in its own tensor language. Covers describing data as tensors, moving them between memories, and running operators on them, plus the pcore programs that define those operators, with complete worked examples. VisionAI Stack Programmer's Guide →How to build vision and AI applications by connecting ready-made processing nodes into a graph. The nodes cover edge detection, colour conversion, resizing, optical flow, TensorFlow models and transformer models for LLM and VLM inference, and custom nodes can be added alongside them. MicroPython Programmer's Guide →How to build those same applications from Python, using the same graph of nodes and tensors. The simplest way to use ztachip: a few lines of Python drive the accelerator at full speed, with no hardware knowledge required, just Python skills.