Use a matrix when a choice has several good options and no obvious winner. First remove options that break a hard limit (must-pass screen). Then pick 4-7 criteria taken from your requirements, give them weights that sum to 100 %, write down what each score 1-5 means for each criterion before you score, and score every option. Weighted total = sum of (weight % x score) / 100, so it ranges from 1.00 to 5.00. Finish with the sensitivity check; if the winner changes, the result is not robust and you need better data or a clearer requirement. Copy the result into the decision record in design-log.md. A spreadsheet version is in decision-matrix.csv.
Choosing criteria:
Decision: D-<nn> <question> Date: <YYYY-MM-DD> By: <...>
An option that fails any line is removed and not scored.
| Hard limit (with requirement or rule ID) | Option A <...> |
Option B <...> |
Option C <...> |
Option D <...> |
|---|---|---|---|---|
<e.g. current rating >= ... A> |
pass / fail | |||
<e.g. in stock at the assembler> |
||||
<e.g. package fits the build method (DFM-3)> |
||||
<...> |
| Criterion (req. ID) | Weight % | 1 | 2 | 3 | 4 | 5 |
|---|---|---|---|---|---|---|
<criterion 1> |
||||||
<criterion 2> |
||||||
<criterion 3> |
||||||
<criterion 4> |
||||||
<criterion 5> |
||||||
<criterion 6> |
||||||
<criterion 7> |
||||||
| Sum | 100 |
| Criterion | Weight % | Option A | Option B | Option C | Option D |
|---|---|---|---|---|---|
<criterion 1> |
|||||
<criterion 2> |
|||||
<criterion 3> |
|||||
<criterion 4> |
|||||
<criterion 5> |
|||||
<criterion 6> |
|||||
<criterion 7> |
|||||
| Weighted total | 100 | ||||
| Rank |
Change the biggest weight by +10 and by -10 points. Spread the opposite change evenly over the other criteria so the sum stays 100 %. Recalculate the totals.
| Case | Weights used | Winner | Winner changed? |
|---|---|---|---|
| Base | - | ||
| Biggest weight +10 | yes / no | ||
| Biggest weight -10 | yes / no |
Result and comment: <winner; robust or not; what data would settle it>
Part of an imaginary 18650 EDC light, > 1000 lm, with a single 3 V class LED. Driver topologies are explained in docs/choosing-parts.md and DRV-1. All weights and scores below are illustrative, chosen to show the method, not measured.
| Hard limit | FET direct drive | Linear regulator | Buck | Boost + 6 V LED | Buck-boost |
|---|---|---|---|---|---|
| Turbo current from one cell possible | pass | pass | pass | pass | pass |
| Flicker-free dimming possible (LED-6) | pass (PWM >= 3 kHz) | pass | pass | pass | pass |
| Parts in stock at the assembler | pass | pass | pass | pass | pass |
All five pass in this example. A real screen often removes one or two.
| Criterion | Weight % | 1 | 2 | 3 | 4 | 5 |
|---|---|---|---|---|---|---|
| Regulation over the discharge (R8, R9) | 25 | Output follows the cell voltage all the time | Regulated only in low modes | Regulated until the cell nears the LED voltage, then falls | Regulated except near the end of the discharge | Flat output from full to LVP cut-off |
| Efficiency in most-used modes (DRV-3) | 20 | Most of the power lost as heat | Poor in most-used modes | Good when the cell is low, poor when full | Good in all modes except a few | Good in all modes |
| Low-mode stability (LED-7) | 15 | Moon flickers or differs unit to unit | Stable only with extra filtering | Stable, small unit-to-unit spread | Current-controlled, small spread | Current-controlled down to the lowest level |
| Cost | 15 | Most expensive of the options | Middle | Cheapest of the options | ||
| Size on the board | 15 | Does not fit without a bigger board | Fits with difficulty | Fits | Fits with room to spare | Fewest, smallest parts |
| Design complexity | 10 | New switching design, layout-critical, loop compensation | Switching design from a reference layout | Some analog design | Few parts, simple layout | Almost nothing to design |
| Sum | 100 |
| Criterion | Weight % | FET direct drive | Linear regulator | Buck | Boost + 6 V LED | Buck-boost |
|---|---|---|---|---|---|---|
| Regulation over the discharge | 25 | 1 | 3 | 3 | 5 | 5 |
| Efficiency | 20 | 4 | 3 | 4 | 4 | 3 |
| Low-mode stability | 15 | 2 | 4 | 3 | 3 | 3 |
| Cost | 15 | 5 | 4 | 3 | 2 | 2 |
| Size | 15 | 5 | 4 | 3 | 2 | 2 |
| Complexity | 10 | 5 | 4 | 3 | 2 | 2 |
| Weighted total | 100 | 3.35 | 3.55 | 3.20 | 3.30 | 3.10 |
| Rank | 2 | 1 | 4 | 3 | 5 |
Example calculation, linear regulator: (25 x 3 + 20 x 3 + 15 x 4 + 15 x 4 + 15 x 4 + 10 x 4) / 100 = 3.55.
The biggest weight is regulation (25 %).
| Case | Weights used (reg / eff / low / cost / size / cplx) | Winner | Winner changed? |
|---|---|---|---|
| Base | 25 / 20 / 15 / 15 / 15 / 10 | Linear regulator (3.55) | - |
| Regulation +10 | 35 / 18 / 13 / 13 / 13 / 8 | Boost + 6 V LED (3.54; linear 3.47) | yes |
| Regulation -10 | 15 / 22 / 17 / 17 / 17 / 12 | FET direct drive (3.67; linear 3.63) | yes |
Result and comment: the winner changes in both directions, so this result is not robust. It depends on how much flat output matters. Go back to the requirement sheet: if a flat output at 30 s and after 10 min (R8, R9) is a Must, the boost or buck-boost wins; if cost and size dominate, the FET or linear driver wins. Record the reasoning in the design log.