Electronics roadmap: from zero electronics (v0) to independently sampled tiles
v0 (this prototype)
| where | electronics |
|---|---|
| helmet | none. The tiles carry only the provisions: simulated routing grooves on the back, a tail land at the −x edge, a 1.5 mm mount gap for a future FPC, and rim exits |
| 3018 placer | one KW10 microswitch on the 3018 board's existing probe header (A5 + GND, $6=0), and a 12 V vacuum pump on its own wall adapter, always on. No board, no firmware, no relay (cnc-placer.md) |
v1: one functional 64-channel tile
The aim is the simplest architecture that scales to 1,024 channels without 1,024 wires leaving the head. Digitise at the tile and send only a digital bus off it.
64 spring-pin contacts ─► comb FPC on the tile back ─► FPC tail (outside the curved area)
─► 2 × 40-way 0.5 mm ZIF (SMT on the tile pod PCB; the tail inserts into it)
─► 64-ch AFE on the pod ─► SPI / LVDS ─► hub (MCU, isolation, battery) ─► USB / BLE
16 tiles = 16 pods on one bus each ─► one hub
| block | choice | why |
|---|---|---|
| contact | gold-plated pogo pin per post, with the ball (gold-plated brass or Ag/AgCl) as its tip; the spring sets a defined contact force | the force limit and compliance belong to the contact itself; a rigid post never touches skin |
| interconnect on the tile | kirigami / comb FPC: a spine along the tile's −x edge with one finger per row of 8. Each finger bends in one direction only, so it follows the curve as a developable strip | copper sees bending strain only (well below 1 % at R > 50 mm for 12 µm Cu on 25 µm PI). A planar FPC thermoformed onto double curvature would need 2–5 % in-plane strain (flat_sheet_strain_pct in layout.json). Rolled-annealed copper fatigues and polyimide does not soften at PETG forming temperatures, so forming FPC is not qualified and is not planned |
| tail and connector | the FPC tail leaves the tile flat, outside the curved region, and plugs into 2 × 40-way 0.5 mm-pitch FPC ZIF (Hirose FH12-40S-0.5SH(55) class: SMT, right angle, bottom contact) on the pod PCB | 80 lines = 64 signal + 2 REF + 2 BIAS + 12 GND/guard. A single ≥ 70-way 0.3 mm ZIF does not exist in a common series (FH35C tops out at 61), so use two standard ones |
| AFE | one 64-channel chip per pod: Intan RHD2164 class (64 unipolar inputs, common reference, lower cutoff settable 0.1–500 Hz, on-chip ADC, SPI/LVDS) | one chip, one bus per tile. Risk: the RHD2164 BGA is listed out of stock indefinitely. Fallbacks: 2 × RHD2132 (QFN) or 8 × ADS1299 (24-bit, 8 ch, more boards and more power) |
| hub | small MCU (ESP32-S3 or nRF52 class) with SPI to the pods, a battery and a charger | reuses the parts and firmware patterns already used in labrat's servo hub |
| 1,024-ch acquisition, bench | Intan RHD recording controller C3008 (1,024 ch; quote only) or an open-source equivalent | the 16 × 64 split matches its port structure |
Reference, bias, shielding. Use one common reference electrode (mastoid or Cz-adjacent contact) and a driven bias / right-leg electrode. Each finger of the comb FPC carries its signal traces between GND guard traces, with a hatched ground plane on the second layer. The pod sits close to the tile so the analog run stays under about 40 mm, and everything after the AFE is digital.
Electrical safety. Battery only, with no mains path while worn. Charging is interlocked: the device does not operate while it charges. If a USB link to a mains-powered PC is used, it goes through a digital isolator. Patient leakage currents and creepage follow IEC 60601-1 principles (aim for type BF) and IEC 60601-2-26 for EEG. This is design guidance, not a compliance claim.
Contact mechanics are a separate engineering problem. Specify the force per contact (dry-electrode practice is roughly 0.5–2 N), pin travel, skin-contact material (gold or Ag/AgCl, plus a biocompatibility review), and validation (impedance at 10 Hz versus force, comfort over hours). None of this is assumed from the v0 geometry.
What more contacts buy, physically
Scalp EEG measures potential differences on the skin. The potentials come from cortical current sources, but they are spread by volume conduction through cerebrospinal fluid, skull and scalp. The low-conductivity skull smears a point source over several centimetres of scalp. Denser contacts improve spatial sampling (they avoid spatial aliasing of the scalp field) and give the inverse problem more constraints for source localisation. Beyond a spacing of roughly 1 cm, the gain comes mostly from better conditioning and from rejecting noise and artefacts, not from new detail.
It does not produce camera-like images of the brain, and it is not radio-frequency or electromagnetic imaging. Future sensing modalities (for example optical fNIRS/DOT, or OPM-MEG) should be judged on their own physics, resolution and depth sensitivity.
Longer term: molded / 3D interconnect research
The goal is to pattern conductors directly on the curved tile, replacing the separate FPC. The figures below are typical ranges from the literature and vendor data, for comparison only. Qualify any process on our own coupons.
| process | min line / space | conductor (typ.) | strain / forming | throughput | cost to start | fit for exocortex |
|---|---|---|---|---|---|---|
| LDS (laser direct structuring) + electroless Cu/Ni/Au | 75–150 µm | electroless Cu 5–15 µm, near bulk | none after plating (rigid part) | high (laser + batch plating) | LDS-grade resin (PC/ABS-LDS, LCP) + plating service | strong: pattern on the moulded curved tile; needs an LDS-grade moulding |
| laser activation / seeding + electroless on standard polymers | 50–100 µm | as LDS | none after plating | medium–high | process development + plating chemistry | good: no special resin; adhesion must be qualified |
| aerosol jet (Ag nanoparticle ink) | 10–30 µm | 2–5× bulk Ag after sintering | prints on the curved part: no forming | low (serial) | high (tool about $200k+), or a service bureau | good for prototypes and fine fan-out; slow for 1,024 lines |
| inkjet (Ag ink) | 30–100 µm | 2–10× bulk Ag | planar or slightly curved | medium | low–medium | planar steps (tail, pod) |
| in-mold electronics (printed stretchable Ag on film, thermoformed, overmoulded) | 100–200 µm | about 15–50 mΩ/sq after forming | designed for 20–50 % forming strain | very high at volume | high tooling | good at volume for routing; line density is limited |
| thermoformable inks (print flat, then form) | 100–200 µm | resistance rises 2–5× after forming | about 10–30 % at forming temperature | high | medium | the honest version of "route flat, then form" |
| multi-material AM (conductive filament / DIW Ag paste) | 0.1–0.5 mm | filament: very high (~0.5–1 Ω·cm); DIW Ag: near ink | prints in 3D | low | low | test fixtures only |
| microvias / embedded conductors (HDI flex) | 50–75 µm lines, 50–100 µm vias | rolled or plated Cu | planar build, then comb-cut to bend | high | standard PCB vendors | v1 baseline (the comb FPC) |
Ordering for the roadmap: comb FPC (v1, buy) → aerosol-jet fan-out on the curved tile (v2, service bureau) → LDS or laser-activated plating on a moulded tile (v3, own process).