exocortex v1 · EXO-410 outline study · solver exo/quad_layout.py

Irregular quadrilateral boards cover 86% of the scalp region, up from 63.5%

Each of the 16 tiles keeps its position, tilt and contact-face height. Only its flat board's outline changes: four edges slide and tilt outward until a real constraint stops them. The 5 mm corner chamfers stay, because the v1 shell was solved to clear chamfered corners and the optical sensors sit in them.

Scalp region under a board

Contacts per helmet

Board area, 16 tiles

Pin length (raw)still the 7.5–12.5 mm pin family

Tightest clearancepin zone to pin zone, need ≥ 0.60

quad board stacksquare boardcontact on the scalp

drag to orbit · scroll or pinch to zoom

What limits each edge

An edge stops moving at the first of these. Every limit is checked in 3D, on the real tile frames.

constraint	rule
neighbouring stacks	board stack (contact face to back parts, 5.5 mm) ≥ 1.0 mm from every neighbour's stack; ≥ 0.6 mm from its pin zone
cuff line	every corner, projected onto the scalp, stays above the beanie cuff line
shell	back parts stay inside the v1 shell's inner face less 0.5 mm, so the shell does not get thicker
hub	every stack ≥ 3 mm from the EXO-430 hub and its REF/BIAS pins
pins	a nut site gets a contact only if its pin is ≤ 13 mm raw (it rounds into the 12.5 mm EXO-5125)

Clearances are separating-axis lower bounds on convex prisms, so the true gaps are at least these. Contacts sit on a 5.3 mm grid, rotated and offset per board to fit the most nut sites.

What it costs

16 different contact boards. Today's EXO-410 is one design built 16 times. Each quadrilateral needs its own placement and its own route, and EXO-410 alone took hours of Freerouting plus HDI vias to route.
16 PCBA setups per order. JLC charges setup and a stencil per design and side, about $68 for a double-sided board, so roughly $1,100 per order before any parts.
The controller may have to change too. In 8 of the 16 tiles (T01, T02, T04, T06, T08, T12, T14, T16) a tilted edge cuts a corner of today's 44.2 mm square, which is where EXO-420's DF40 connectors sit. Keeping one controller design means pinning those edges so they never cut into the square; the solver can do that at some cost in coverage.
The shell changes: standoffs, FPC slots and splice tabs move with each board. The shell's thickness does not, because that was a constraint.

Per tile

tile	region	area vs square	contacts	corner angles (°)	pin raw (mm)

Your question

Would flex PCBs work?

A flat sheet can bend around one axis freely, but the head curves in two directions at once. Pulling a tile-sized sheet onto it means stretching the sheet in its own plane, and copper on polyimide is made to bend, not stretch. The layout already measures this for every tile: 1.4–2.0 % of in-plane strain for a 44 mm sheet. That is far past what copper traces survive, so a whole-tile flex board would wrinkle or crack.

option	coverage	contacts sit	designs	main problem	verdict
Rigid squares (today)	63.5 %	flat board, pins 7.5–12.5 mm	1	corners sit 6.4 mm farther from the scalp than the centre, so the helmet is about 12 mm thicker	bench order
Rigid irregular quads (this page)	86 %	same	16 contact boards, maybe 16 controllers	16 routes and 16 PCBA setups per order	possible
One flex sheet per tile	high	on the curve	1	needs 1.4–2 % stretch; and the board needs 6 layers with HDI vias, which is specialist flex work	no
Slit / petal flex (kirigami)	high	on the curve	1	every pin needs a nut, a follower within 2 mm and a stiffener, which leaves little bendable flex; the converters and their 0.4 mm pins still need a rigid, multilayer spot	hard
Rigid-flex islands	≈ as quads or better	near the curve	1	rigid-flex fabrication and folded assembly cost more; the islands still need fine-pitch multilayer routing	best v2 candidate

Rigid-flex islands, sized by what the board already is

One small rigid island per ADS131M08: its 2 × 4 contacts, their followers and the converter, about 16 × 27 mm. Flex hinges join the 8 islands of a tile and let each one tilt to follow the head.
Sag scales with the square of the size. An island's diagonal is about half a tile's, so the gap between board and scalp shrinks from about 6 mm at a tile's corners to roughly 1.3–1.6 mm (an estimate from the layout's tile forming depths of 5.3–6.5 mm).
That would cut the six pin lengths to one or two and make the helmet about 5 mm thinner. Every island would be the same part, so it stays one design, unlike the irregular quads.
Packing gets better too: small islands follow the surface, so the gaps between tiles close the same way they do on this page.

Recommendation: keep the routed square boards for the 1–2 tile bench order. Take irregular quads only if the gain is wanted before a redesign. Treat rigid-flex islands as the v2 study, and get a rigid-flex quote and a one-island prototype before changing the architecture.

Inputs: out/layout.json (tile frames), out/pcb/tile_geom.json (contact-face heights, pin table), out/shell_v1_report.json (shell inner face 23.2 mm). Output: out/quad_layout.json. Generated 2026-10-27.