From Prototype to Mass Production: Sensor Manufacturing

We have working sensors in the lab. Now we need to make 100,000 of them. The challenges are completely different.

Prototype vs Production Mindset

In prototyping:

• Hand-tuned calibration
• Best-in-class components
• Unlimited debug time
• Sample size: 1

In production:

• Automated calibration (seconds per unit)
• Cost-optimized components
• Diagnosis in seconds
• Every unit must work

Yield: The Defining Metric

Yield = (Good units) / (Total units manufactured)

At $50 component cost and 80% yield:

• Effective cost = $50 / 0.8 = $62.50 per good unit
• 20% scrap cost adds up to millions at scale

At 95% yield:

• Effective cost = $50 / 0.95 = $52.63 per good unit

That 15% yield improvement might be worth more than any other optimization.

Sources of Yield Loss

Component variation: Every resistor, capacitor, optical element has tolerance

• Resistors: ±1% is common
• Optical elements: ±5% on transmission is good
• Combined: distributions multiply

Assembly variation: Placement accuracy, bond quality, contamination

• Pick-and-place: ±50μm typical
• Die attach: thermal interface quality varies
• Cleanliness: particles in optical path

Process variation: Temperature, humidity, equipment drift

• Reflow profiles vary oven-to-oven
• Adhesive cure varies with batch

Design for Manufacturing (DFM)

Choices we made for manufacturability:

Wider tolerances where possible: If algorithm can handle ±10% depth error, don't specify ±5% sensor calibration.

Testability: Every subsystem must be independently testable. Can't debug what you can't probe.

Redundancy: If one sensor fails calibration, can we reroute to a spare? Built-in flexibility.

Binary pass/fail: Clear criteria, no subjective decisions on the line.

Calibration at Scale

Our lab calibration takes 30 minutes per unit. Production target: 30 seconds.

How:

• Parallel calibration: Calibrate multiple parameters simultaneously
• Known-good reference: Calibrate against golden units, not absolute standards
• Statistical shortcuts: Some parameters can be inferred from others
• Post-assembly trimming: Adjust in firmware what hardware can't deliver

Quality Escapes

What happens when a bad unit reaches a customer?

• Field failure → customer frustration → returns → reputation damage
• Cost of one field failure = 100x cost of factory catch

Invest heavily in outgoing quality control (OQC):

• 100% functional test
• Sample environmental stress screening
• Accelerated life testing on samples

Learning from the Line

Production data is gold:

• Calibration parameters → track component suppliers
• Test failures → feed back to design
• Customer returns → find escapes in testing

We've built dashboards tracking key metrics in real-time. Manufacturing problems visible within hours, not weeks.

Production ramp is Q3. We'll learn a lot more soon.