Retrofittable Embedded Vision and Energy-Efficient FPGA Inference for Production and Re-manufacturing Lines

Kuantek Elektronik A.Ş. is a Turkish deep-tech SME based at Bilişim Vadisi Technology Park (Gebze), joining consortia as the embedded vision and edge inference partner.

Disassembly and re-manufacturing cells face products of unknown condition and configuration, so automation depends on recognising the part and assessing its state at line rate. PV production beyond the state of the art needs inline thermal and electroluminescence inspection that keeps pace with throughput. Both problems have the same constraint: the vision has to run next to the machine, on hardware that can be retrofitted to an existing cell, without a GPU rack behind every station.

Our contribution: quantised CNN inference accelerators implemented in RTL on FPGA, with the power and precision profile adapted to the workload instead of fixed; compact sensing nodes that attach to existing equipment; thermal and visible-band imaging, including an existing thermal inspection platform available as a background asset for adaptation within the project; and native integration over Modbus, CAN and EtherCAT. Energy per inference is a measurable figure we can report as a project KPI.

We can lead a hardware or edge-processing work package, supply demonstrator nodes for pilot lines, and act as a technology supplier within cascade funding (FSTP) schemes. We are looking for coordinators, production technology partners and research organisations needing an industrial partner able to turn algorithms into deployable hardware.