Thermal Budget — First Pass
Dissipation estimates per LED candidate and driver, and what they imply for PCB copper vs. MCPCB vs. discrete heatsink. Enclosure thermal design is a later phase.
This is a first pass, sized to inform the LED/driver choice now, not a final thermal design — the enclosure is explicitly out of scope for this epic. The goal here is narrower: which LED candidates from led-candidates.mdx are compatible with plain FR-4 copper, and which force an aluminum MCPCB or discrete heatsink onto the board split before the architecture pass locks anything in.
Method
Two dissipation sources on the LED board:
LED junction dissipation — for an LED running at forward voltage Vf and current I, essentially all of Vf×I becomes heat (a few percent leaves as light — commercial white LEDs are roughly 20–40% wall-plug efficient at the system level, but for a conservative thermal budget it's standard practice to treat the full electrical input as heat).
Driver dissipation — a buck converter isn't 100% efficient; the undelivered fraction becomes heat in the driver IC, inductor, and catch diode. For the driver candidates in
driver-ics.mdx(PT4115, AL8860, TPS92511 — all simple non-synchronous bucks in this power class), a working assumption of 85–92% efficiency at these current levels is reasonable pending bench measurement; this pass uses 88% as a mid-estimate.
Driver dissipation ≈ P_LED × (1/η − 1). At η=0.88, that's roughly 14% of the delivered LED power showing up as driver-side heat — e.g. a 3.6W LED load implies ~0.5W dissipated across the driver IC + inductor + diode, concentrated mostly in the IC package.
Per-candidate dissipation
| LED candidate | Per-unit dissipation | Notes |
|---|---|---|
| SMD2835 (60mA bin) | ~0.2W | e.g. XL-2835UWC-02 (C2843876) — Vf 3.4V × 60mA |
| SMD2835 (150mA bin) | ~0.6W | e.g. HL-A-2835D46W (C210329) — Vf 3.4V × 150mA |
| SMD5730 (150mA) | ~0.5–0.6W | all three 5730 candidates cluster here |
| CXA1304 COB (9V/400mA bin) | 3.6W in one 13.4×13.4mm package | concentrated — see below |
A discrete-array design multiplies the per-unit number by however many LEDs the architecture pass lands on (parallel strings × series count) — total board dissipation for, say, 12 SMD2835 units at 150mA is ~7.2W spread across the board footprint, very different from putting the same 7.2W into two COB packages.
Driver dissipation estimate
| Driver | Target LED load | Estimated driver-side dissipation (η≈88%) |
|---|---|---|
| PT4115 / AL8860, ~350mA output | ~1–1.2W (a 3s or 4s discrete string) | ~0.14–0.2W |
| PT4115 / AL8860, ~1A output | ~3–4W (higher-current string) | ~0.5W |
| TPS92511, 500mA output | up to ~1.5–2W (single 3s/4s string at its max current) | ~0.2–0.3W |
| One driver per COB module (9V/400mA) | 3.6W | ~0.5W, concentrated in the SO-8-EP/MSOP package |
These are single-digit-tenths-of-a-watt numbers per driver — small next to the LED dissipation itself, but not negligible if several driver ICs are crowded onto a small board (one-driver-per-parallel-string architecture, see led-candidates.mdx).
PCB thermal strategy: copper pour vs. MCPCB vs. discrete heatsink
Rule-of-thumb sizing (standard 2oz copper on FR-4, natural convection, no forced air — consistent with a decorative lamp, not a fan-cooled fixture):
Plain FR-4 with a generous copper pour (thermal vias to a ground/heatsink-plane pour) comfortably handles on the order of ~1W per LED site before local temperature rise becomes a real concern for LED lifetime. This covers every 2835/5730 discrete candidate above (0.2–0.6W each) with margin, provided the pour area per pad is generous and via-stitched — a token pad with no pour will not hit this.
Once a single site needs to shed more than roughly 1–1.5W, plain FR-4 copper starts to struggle — thermal resistance from junction to ambient via a thin FR-4 dielectric is fundamentally worse than a metal-core path. This is exactly the CXA1304 COB case at 3.6W in one 13.4mm package: plain FR-4 copper pour is not a credible thermal path for that candidate. It needs either an aluminum MCPCB (metal-core PCB, thermal resistance roughly an order of magnitude better than FR-4 for the same copper weight) or a discrete heatsink bonded to the COB's thermal pad.
A discrete-array design that totals several watts across many small LEDs (e.g. 12× SMD2835 at 150mA ≈ 7.2W board-wide) sits in between: no single site is hot, but the board as a whole needs enough total copper area and, likely, thermal relief to whatever the LED board mounts to (not just local pours per pad) — this is a board-level layout constraint, not a single-component one.
| LED path | Per-site dissipation | Recommended PCB strategy |
|---|---|---|
| SMD2835 / SMD5730 array | 0.2–0.6W each | Plain FR-4, generous copper pour + thermal vias per pad |
| CXA1304 COB (9V bin) | 3.6W, one site | Aluminum MCPCB or discrete heatsink — plain FR-4 copper is undersized |
Questions the architecture pass must answer
If the discrete-array path is chosen, what's the actual LED count/total board dissipation once the brightness target is fixed — does it stay under the "many small sites, generous pour" regime, or does density push toward MCPCB anyway?
If the COB path is chosen, does the board split put the COB on its own MCPCB daughter-board, or does the whole LED board go aluminum-core (cost/complexity tradeoff, not resolved here)?
Does the driver IC(s) need their own thermal relief beyond a standard pour, given they're clustered with LED heat sources on the same small board?
The 85–92% efficiency assumption above is a first-pass estimate, not a datasheet-sourced number for any specific driver at the specific current the architecture pass will pick — worth bench-confirming once a driver is selected, per the general "not yet bench-confirmed" caveat already flagged for the PD front end in issue #5.