This program improved an ophthalmic laser device, focusing on laser-output stability and RFID consumable management. We also reworked production and test flows and introduced semi-automated testing and analysis to improve quality and traceability.
Recovering the architecture through reverse engineering
The product was already shipping but suffered yield problems. Its original R&D team had dispersed, design documents were outdated, and several conflicting PCB Gerber versions existed. I spent one month at the customer factory reverse-engineering the hardware and firmware and correcting the Cable Assembly / Pin-out, Power Tree, and PCB Functional Block Diagram.
I redefined responsibilities across multiple PCBs, MCU roles, and inter-MCU protocols, and improved board-to-board wiring without changing mechanical constraints. Using KiCad, I created the first schematic revision from legacy circuits and new requirements for the hardware team to refine. I also corrected the BOM and aligned it with the factory production-management system.
After the new boards were built, I spent another three months on site validating each board. We resolved short-circuit risk from incorrect connector definitions, cooling-fan failures caused by mismatched peripheral specifications, and unusable laser-energy feedback caused by sensing and trace-layout problems. Two board revisions substantially improved completeness and stability.
Rebuilding the firmware environment
The original TI TMS320 + Parallax P8X32A design depended on CCS 5.5 running on one Windows 7 machine, while the P8X32A used Spin. I migrated the architecture to TI TMS320 + STM32G431, upgraded to CCS 10 and STM32CubeIDE, standardized on C/C++, and rebuilt the workspace and MCU project skeletons.
I added Windows, macOS, and Linux development support and Git version control, replanned the TMS320 pin-out, and converted legacy settings to CCS10 SysConfig. I also fixed a boot failure caused by the _c_int00 entry point odd address, restoring reliable firmware startup.
RFID consumable management
I ported the RFID protocol from P8X32A to STM32G431 and tuned parameters so ENDO tags were readable only within a defined distance window, reducing false reads from other consumables. I also established a pre-shipment validation flow for factory-built RFID antenna coils.
Laser-energy sensing and control
The original sensing circuit and missing firmware accumulation logic could not provide stable feedback. I adapted a previous-generation design to cover four wavelengths and a 0.5–200 mW range, drafted the first schematic revision, and worked with the hardware team through final PCB design.
Firmware used the STM32G431 12-bit ADC with DMA for stable, high-throughput acquisition. Dual OpAmp gain paths covered low- and high-power ranges, while an energy-accumulation algorithm produced usable closed-loop feedback and significantly improved laser-output stability.