cd ~/
HardwareSpecificationDecision MakingLeadershipMagic Leap

Lessons from Spec Freeze: Hardware-Software Co-Design

What I learned from our first major hardware spec freeze - the art of committing to specifications when uncertainty remains.

Evyatar Bluzer
2 min read

We've hit spec freeze for the perception system. This means the sensor selections, interfaces, and hardware architecture are locked. Software can still evolve, but the physical substrate is committed.

It's terrifying. And necessary.

The Spec Freeze Dilemma

Hardware has long lead times:

  • Custom sensor development: 18-24 months
  • ASIC design: 12-18 months
  • Tooling for manufacturing: 6-12 months

Software timelines are:

  • Algorithm improvement: days to weeks
  • Integration: weeks to months

This mismatch creates a fundamental tension. We must freeze hardware while software is still immature.

What We Committed To

Depth sensor: ToF at 320x240, 30Hz, 0.3-5m range

  • Rationale: Better outdoor performance than structured light
  • Risk: Lower resolution may limit hand tracking quality

Tracking cameras: Global shutter, 640x480, 60Hz

  • Rationale: Global shutter essential for fast motion
  • Risk: Power consumption higher than rolling shutter

Processing: ARM + DSP + custom accelerators

  • Rationale: Best power efficiency for known workloads
  • Risk: Less flexibility than pure GPU approach

Decision Framework

For each decision:

  1. Identify irreversibility: What can't be changed later?
  2. Quantify uncertainty: What don't we know?
  3. Define escape hatches: What flexibility can we preserve?
  4. Accept residual risk: What's the worst case?

Mistakes I'm Worried About

Being honest with myself:

  • Depth resolution: 320x240 may not be enough. Software can compensate somewhat through temporal filtering and super-resolution, but there's a floor.
  • No dedicated ML accelerator: Betting we can do most inference on DSP/GPU. If neural networks become central to tracking, we'll be power-constrained.
  • Eye tracking camera resolution: Balanced against power, but may limit pupil tracking accuracy in bright conditions.

Key Learnings

  1. Get algorithms to 80% before freeze: You need enough fidelity to make informed hardware choices.

  2. Build margins: Specified 20% power margin, 30% compute margin. We'll use it.

  3. Prototype the risky bets: Our structured light experiment de-risked the ToF decision.

  4. Document assumptions: Every spec has underlying assumptions. Write them down. When assumptions change, you know which specs to revisit.

  5. The spec is a contract, not a wish: If it's in the spec, it's a commitment. Don't spec what you can't deliver.

2017 will be about delivering on these commitments. Onward.

Comments