Sundance DSP's TinyBeast FPGA Is a Compact Microchip PolarFire FPGA Board for Edge AI Acceleration
Compact boards aim to deliver an easy route to custom logic and edge AI acceleration through Microchip's VectorBlox.
Edge computing specialist Sundance DSP is preparing to launch a compact yet powerful field-programmable gate array (FPGA) device based on Microchip's PolarFire platform β available in mini-PCI Express (mPCIe) card and surface-mount module variants and delivering full support for Microchip's VectorBlox artificial intelligence (AI) acceleration: the TinyBeast FPGA.
"TinyBeast FPGA is a compact new processing solution that can speed up your industrial or embedded computer," Sundance DSP claims of its latest device design. "Leverage the power of PolarFire FPGA technology to offload demanding tasks from your central processor for a smoother and significantly faster system experience. This translates to ideal real-time performance in industrial applications that require efficient data handling. Furthermore, its compact package opens doors for innovative solutions in space-constrained environments."
The two TinyBeast FPGA boards β one designed to connect to the mPCIe socket on an existing motherboard, the other with high-density connectors on its underside for installation on a custom carrier board as a surface-mount module β are built around Microchip's PolarFire FPGA MPF300T-1FCVG484E, giving them 300k logic elements, 924 18Γ18 math blocks, 20.6Mb (around 2.5MB) of total RAM, 16 SERDES lanes, two PCI Express endpoints of which one is exposed on these boards, and up to 512 user-accessible input/output (IO) pins.
While that's plenty of flexibility in itself, Sundance DSP is making much of the TinyBeast's potential for accelerating on-device edge artificial intelligence (edge AI) and machine learning workloads using Microchip's VectorBlox β a technology we tested in our FPGAdventures series using the PolarFire SoC Video Kit. "VectorBlox is a hardware acceleration engine specifically designed for efficient AI inferencing," the company explains. "This translates to real-time processing of complex algorithms directly on the TinyBeast FPGA, enabling applications like predictive maintenance, anomaly detection, and image recognition directly at the industrial edge."
Sundance DSP has promised to release full schematics for both the mPCIe and system-on-module variants of the boards β dubbed the TinyBeast FPGA P and S respectively β under an as-yet unspecified open license, along with sample firmware and an example PC package for interfacing with the FPGA. "We aim to sustain a vibrant developer community around TinyBeast FPGA by providing these resources," the company says, "which we will release this summer."
Those interested in learning more can sign up to be notified when Sundance DSP's crowdfunding begins on the project's Crowd Supply campaign page.