Circuit Valley CHC5 Hits 2000 FPS on an FPGA Camera

A Machine Vision Camera You Can Actually Take Apart

Industrial cameras are usually sealed appliances: fixed sensor, fixed interface, fixed firmware, and a support contract. Circuit Valley's CHC5, which went live on Kickstarter and was detailed by CNX Software on September 2, 2026, takes the opposite approach. It splits the camera into a core board, an interface board, a sensor board and an enclosure, and lets you swap each one.

• Core: AMD Xilinx Zynq-7020 SoC FPGA with dual-core Arm Cortex-A9, 1 GB DDR3, 1 Gbit flash and a microSD slot
• FPGA fabric: 85K logic cells, 4.9 MB block RAM and 220 DSP slices available for custom image processing
• Throughput: a 600 megapixel/second ISP, giving 2000 FPS at 640x480, 170 FPS at 1080p and 40 FPS at 4K
• Price: $684 for the base system including core PCB, interface PCB, CNC enclosure, sensor and mount, with delivery expected January 2027

What Does an FPGA Buy You in a Camera?

The Zynq-7020 is a hybrid part: two Cortex-A9 application cores sitting next to programmable logic. In the CHC5 that split maps cleanly onto the workload. The Arm cores run Linux and handle the things software is good at — configuration, networking, protocol handling. The FPGA fabric handles the per-pixel work that would drown a CPU: correction, filtering, and any custom machine vision or AI pipeline you care to synthesise.

That matters most where latency is the requirement rather than throughput. A pick-and-place machine or a high-speed inspection line needs a decision in microseconds after the shutter, not after a round trip through a host PC. Doing that work in fabric is why 2000 FPS at VGA resolution is achievable at all.

It is the same argument that keeps drawing makers back to programmable logic. We covered it when the Forgix board paired an RP2354 MCU with an FPGA for $50, and the CHC5 is that idea scaled up to a complete imaging product.

Sensors, Mounts and Interfaces You Choose

The sensor board accepts parts using MIPI D-PHY, LVDS, Sub-LVDS, Low Voltage LVDS or SLVS interfaces, with support for sensors up to 128 MP. Lens mounts on offer include C-Mount, CS-Mount, Canon RF, Sony E, M42 with multiple flange distances, S Mount, M43 and custom options, down to a minimum 8.75 mm flange distance.

The interface board decides how images leave the camera: 10 Gbps or 5 Gbps USB 3.0, Gigabit Ethernet with GigE Vision protocol support, optional PoE, and optional HDMI. Multiple outputs can run simultaneously, which is genuinely useful when you want a live HDMI feed for a human operator alongside a GigE Vision stream feeding an inspection pipeline.

Who Is the CHC5 For?

At $684 for a base configuration, this is not competing with a $30 Raspberry Pi camera module, and it is not trying to. It sits in the gap between hobbyist imaging and five-figure industrial systems — research labs, small manufacturing lines, robotics teams and anyone who has hit the ceiling of what a $300 on-device AI camera like the reCamera Pro can do and does not want to jump straight to a sealed commercial unit.

Two caveats worth stating plainly. This is a crowdfunding campaign with a modest €10,000 target and rewards scheduled for January 2027, so the usual crowdfunding timing risk applies. And while the project brands itself around openness, buyers should confirm exactly which parts of the design and toolchain are published before committing — that detail matters more on an FPGA product than almost anywhere else, because the value of programmable fabric depends entirely on being able to program it.

For readers tracking the broader trend, this is another data point in FPGAs quietly returning to the small computing conversation as edge AI workloads outgrow what a general-purpose CPU can do inside a power budget.

Sources: CNX Software — September 2, 2026; CircuitValley Kickstarter campaign — September 2026; CircuitValley project write-up — August 2026.