Shypd.aiThe largest open index of AI toolstinyTPU is an open-source hardware implementation of a Tensor Processing Unit (TPU) designed for embedded systems and IoT. It offers a resource-friendly machine learning co-processor that can be scaled for various FPGA sizes.
Trending
About
tinyTPU is an open-source project providing a hardware implementation of a Tensor Processing Unit (TPU) specifically tailored for embedded systems and IoT devices. Unlike Google's original TPU, tinyTPU focuses on fixed-point arithmetic and is designed to be resource-friendly, allowing it to be deployed across different FPGA sizes, from small embedded systems to larger data centers. Its architecture includes a Weight Buffer, Unified Buffer, Systolic Data Setup, a Matrix Multiply Unit (MXU), Accumulators, and Activation functions (Sigmoid and ReLU). The system is configurable, allowing customization of component sizes. It communicates via an AXI interface and has been evaluated on the Xilinx Zynq 7020 SoC, demonstrating competitive performance compared to traditional processors.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
