CRYSTALSIM

initializing lattice

Optimization Strategies

Lesson 3 of 6·20 min read·+80 XP
Back to Track

Optimization is just science

Engineers who 'just tweak things until it works' are slower and worse than engineers who follow a method. The method below is borrowed straight from the scientific method and applied to chip design. Five steps. Repeat until done.

The five steps

1. Identify the weakest metric — sort by distance from target, not by absolute number. 2. Determine the root cause — material? geometry? coupling efficiency? 3. Generate 2–4 hypotheses — each one a single, testable change. 4. Test each hypothesis independently — one change, one re-test, record the delta. 5. Combine the winners — re-test the composite. (Sometimes wins don't add — that's information too.)

Key Concept
Controlled variable
The single thing you change between two measurements. If you change two things at once, you cannot attribute the result to either. Always vary one variable at a time.

Practice — fill in the loop

Use the worksheet below to walk a real design through the loop. Pick any test result you don't like, and complete each step. The widget marks the loop complete only when every step has substantive content.

Interactive · Optimization worksheet
Optimization loop
0/5 complete
1 · Identify the weakest metric
Run the full test suite, sort metrics by 'distance from target'.
2 · Determine root cause
Material, geometry, or coupling? Pick one.
3 · Generate hypotheses
List 2–4 testable changes. Don't combine them yet.
4 · Test each hypothesis independently
One change at a time, re-run the same test, record the delta.
5 · Combine the winners
Compose only the changes that helped. Re-test the whole stack.
Fill in each step (≥4 chars) to mark the loop complete.

Exercise — fix a deliberately bad design

Open the Chip Builder. Build (or load via Templates if available) this deliberately weak design:

- Substrate: Silicon - Channel: Silicon (low mobility for our target) - Gate: Metal Gate on SiO₂ (no piezo) - No GAA wrapper - No EM antenna - Wide channel, thick oxide

Run the Switching Speed and Power tests. They will be poor. Apply the 5-step loop until every metric is green. Use the widget above to track your steps. Target: switching speed ≥ 1.5× the starting value AND power ≤ 0.7× the starting value.

Checkpoint · +5 XP
Why change only one variable at a time?
Lesson Summary
  • Apply the scientific method to chip optimization.
  • Five steps: identify weakest → root cause → hypotheses → test → combine.
  • Change one variable at a time. Re-test after each change.
  • Take a deliberately bad design and make every metric green.
Test Your Knowledge · +80 XP
1
Step 1 of the optimization loop is:
2
When testing hypotheses, you should change…
3
If two changes win individually but not together, that's…
4
Root-cause analysis tries to identify whether the issue is:
5
Why follow a fixed loop instead of intuition-driven tweaks?