High-Brightness LED Candidates

Series/parallel white-LED options for the 15V/3A rail, with voltage-headroom and current-sharing analysis. Final pick is an architecture-phase decision.

The LED power stage sits downstream of the PD board ("Board P"), which delivers a switched 15V rail contracted to 3A (45W ceiling). This page catalogs LED options against that rail — it does not pick one. All candidates were verified against the local JLCPCB parts DB (fetched 2026-07-25); per-part lookup dates are noted where they differ.

Candidate 1 — SMD2835 discrete white LEDs (array)

Small (2.8×3.5mm), cheap, extremely well-stocked package. Individually low-power (0.06–0.6W), meant to be used in banks — series strings for voltage, parallel strings for total lumens.

All three lookup-verified, DB date 2026-06-24 to 2026-06-27. Package: SMD2835-2P.

Candidate 2 — SMD5730 discrete white LEDs (array)

Larger die/pad (5.7×3.0mm) than 2835 — more copper contact area per emitter for the same rough Vf/current class, i.e. an easier thermal path per LED at the cost of a bigger footprint. Also array-oriented.

All three lookup-verified, DB date 2026-06-24. Package: SMD5730-3P.

Candidate 3 — COB module (single high-flux source)

A COB puts multiple die on one substrate behind a single pair of pads, avoiding the parallel-string current-sharing problem entirely (see below) at the cost of a much more concentrated thermal load.

CXA1304 9V/400mA family — fully-specified, single-die-equivalent COB in the 3.6W class:

Lookup-verified, DB date 2026-06-27. Package: SMD 13.4×13.4mm.

:::warning All three CXA1304 SKUs show zero current stock in this DB snapshot. JLCPCB restocks COB parts irregularly — treat this as a real sourcing risk to flag for the architecture pass, not just a note. Re-verify at order time regardless of what's chosen.

Also note: this same CXA1304 family is sold in 18V/200mA and 36V/~lm-matched bins (e.g. C510500, C510507 — not tabulated above) that are electrically incompatible with a straight buck topology on a 15V rail (see headroom analysis below). Only the 9V bin is usable here without a boost or buck-boost stage — picking the wrong voltage bin off a parametric search would be a real mistake. :::

Alternate (currently in stock, specs unverified): HL-LM002H384W-7B1C18 (C22398559), SMD 13.5×13.5mm, 3000K, Stock 108, partdetail. Lookup-verified (DB date 2026-05-23) for existence/stock only — its datasheet is a scanned/image PDF that did not yield machine-readable Vf/current/lumen data in this research pass. Usable as a stock-available fallback, but Vf and thermal specs must be confirmed from the manufacturer directly before any headroom or thermal math is done against it.

Voltage headroom analysis

A buck constant-current driver needs the LED string's total forward voltage to sit comfortably below the 15V rail — the difference is the driver's working headroom (dropout + ripple margin). Too little headroom and the driver can't regulate cleanly at the top of the rail's tolerance; too much and efficiency suffers (all binned as switching loss dropping the extra volts).

This mirrors the source issue's own worked example (4s white ≈ 12–13V, leaving ~2–3V for the driver) — the 4s point for 2835/5730 lands right at that edge.

Current-sharing considerations for parallel strings

To reach meaningful total lumen output, one series string (of either discrete package) is not enough — the architecture will need multiple strings in parallel. This is the risk the issue calls out explicitly:

• LED forward voltage is binned, not exact — even LEDs from the same reel can differ by ±0.1–0.2V at a given current. Tie multiple strings directly across the same regulated-current output and the lowest-Vf string hogs a disproportionate share of the current.

• Some LED chemistries have a negative Vf-vs-temperature coefficient at constant current: the hottest string's Vf drops further, letting it draw even more current, which heats it further — a slow thermal-runaway spiral in the worst case (more relevant to high-current single strings than these ~150mA-class discretes, but worth flagging).

• Ballast options, roughly cheapest to most robust:

  1. Per-string ballast resistor — a few ohms in series with each parallel string, sized to drop enough voltage that Vf mismatch becomes a small fraction of the string's total impedance. Cheap, lossy (pure resistive dissipation, subtracts from the LED headroom budget above), simplest to lay out.

  2. Factory-matched/binned strings — buy LEDs in a tight Vf bin and don't rely on ballast at all. Reduces BOM cost of ballast parts but adds a procurement constraint (bin availability, reorder risk) that the JLCPCB catalog snapshot alone can't confirm.

  3. One CC driver output per string — no shared node, no sharing problem, at the cost of one driver IC (+ its full passive set, see driver-ics.mdx) per string. Most robust, most parts.

• The COB candidate (single module) sidesteps this entirely — it's one part, one current path, no parallel-string question. That's a real point in its favor if the sourcing-stock risk above can be resolved.

Thermal path per candidate (see thermal-budget.mdx for numbers)

• 2835/5730 discretes: low per-LED dissipation (0.06–0.6W), but a bright design needs many of them — total board dissipation adds up even though no single site is hot.

• COB: all the heat lands in one 13.4mm² package (3.6W in the CXA1304 9V bin) — copper pour alone is unlikely to be enough; see the thermal page for what's actually needed.

Questions the architecture pass must answer

• Discrete array (2835/5730, many small strings) vs. single COB (concentrated, simpler current path) — which trades better against the enclosure/thermal plan and the target lumen output?

• If discrete: 3s (safe headroom, lower efficiency) or 4s (tight headroom, higher efficiency) per string, and how many strings in parallel?

• If discrete + parallel strings: ballast resistor, Vf-binned procurement, or one driver per string?

• If COB: is the 9V/400mA CXA1304 bin's zero-stock status acceptable to design around (pre-order lead time) or does it force a fallback to the stock-available-but-spec-unverified LM002H384W part (requiring its datasheet be tracked down through the manufacturer, not JLCPCB)?

• What CCT (warm/neutral/cool) does the fire/wave-like modulation concept actually want? Some candidates above only exist in a subset of CCTs at high stock.