The problem
A lidar sensor is an unforgiving thing to build: optical, mechanical, and electronic subsystems all have to come together within tight tolerances, and a small alignment error upstream shows up as a failed unit at final test. The challenge is holding that precision while building enough volume to matter.
I worked on the manufacturing and NPI side of that equation — bringing new builds into production and supporting the running line so good design didn't get lost in the transfer to the floor.
What I did
NPI & design-for-manufacturing
Supported the introduction of new sensor builds — feeding manufacturability and assembly feedback back to design, defining assembly and alignment flows, and standing up the processes a new product needs before ramp.
Optical-mechanical assembly & alignment
Developed and tightened the assembly and alignment steps where most of the yield risk lived, building the fixtures and procedures that made a precise operation repeatable across operators.
The D-Lens attach: two and a half minutes down to twenty-four seconds
Lens attach was the operation that set the pace of the whole build. It ran about two and a half minutes per lens, and a single sensor takes 128 of them — over five hours of touch time before a unit was even complete. Reworking the fixturing and the method brought it to 24 seconds a lens, which turns those five hours into roughly fifty minutes. The gain came from making the precise thing the easy thing to do, not from asking anyone to work faster.
Production support & defect reduction
Diagnosed line and test failures, traced them to root cause, and drove corrective actions to lift first-pass yield and reduce rework. Two examples of the shape that work takes: consolidating epoxies across the line took $40k of scrap out per quarter, and a persistent QFN solder-voiding problem in SMT went away by moving those assemblies to vacuum reflow rather than by tightening an inspection.
Transfer to offshore contract manufacturers
Led the transfer of multiple lidar subsystem assemblies to offshore CMs, scaling optical bench component-attach while holding yield and output targets — and built the active-alignment lens and precision mirror attach processes on FiconTEC platforms that made that coupling performance repeatable somewhere other than the room it was developed in.
The takeaway
When tolerances are tight, yield is a process problem long before it's a people problem. The leverage was in designing the alignment and assembly steps so the right outcome was the easy one — then proving it with data at final test.