Optical Chromatic Memristor Neuron
A buildable DIY guide — a synapse whose weight is written by light, and whose inputs are encoded by color. For ALMAWARE / Eran neuromorphic R&D. Built for Iddo, 2026-06-24. Honest tiers: working-today → real optical memory → chromatic.
1 · The idea in one breath
A neuron computes y = f( Σ wᵢ·xᵢ ) — a weighted sum of inputs, then a threshold. We make it optical + chromatic:
- Inputs xᵢ = light intensity, each input on its own color (Red / Green / Blue… = "chromatic channels", like WDM in fiber).
- Weights wᵢ = light-programmable conductances (the "optical memristor") — a cell whose resistance is set by light and remembered.
- Σ (sum) = photocurrents add up automatically at one wire (Kirchhoff's current law).
- f (fire) = a threshold/comparator (or a transistor for a soft sigmoid).
So: colored light in → each color weighted by its own light-set memristor → currents sum → neuron fires if the sum crosses threshold. Color is the address; light is the signal; the memristor is the memory.
R LED ─▶[red filter]─▶( w_R )─┐
G LED ─▶[grn filter]─▶( w_G )─┤ Σ photocurrents ┌─ threshold ─┐
B LED ─▶[blu filter]─▶( w_B )─┼──────▶ [transimpedance op-amp] ──▶ [comparator] ──▶ OUT (fire / no-fire)
│
(each (w) = a light-set optical-memristor / photoconductor cell)
2 · The honest tier ladder
Each tier is a real, finished milestone. Climb only as far as you want — Tier 0 already works and teaches the architecture.
TIER 0 Optical-chromatic perceptron — works today, ~$25, no exotic materials
Weights = LDR photoresistors behind color filters. Not a true memristor (no long-term memory — you set weights by hand), but a genuine optical, chromatic, analog weighted-sum-and-fire neuron. This is your proof-of-concept and test rig for everything above it.
TIER 1 Real optical MEMRISTOR synapse — Persistent Photoconductivity (ZnO)
Replace the LDR with a cell that remembers. ZnO (and some oxides/perovskites) show Persistent Photoconductivity (PPC): a blue/UV pulse raises its conductance and it stays up for minutes–hours in the dark = an analog, multi-level, light-written weight. Reset with heat or red/IR. This is the genuine "optical memristor / optoelectronic synapse" — well documented in the literature, DIY-able with care.
TIER 2 Chromatic multiplexing — the prism / cut-stone
One white input beam through a prism or cut crystal fans into a rainbow; each color lands on a wavelength-matched synapse. Now a single beam's spectrum carries many weighted inputs at once — true "chromatic weights" (your cut-stone idea, made literal).
TIER 3 Array + optical learning
Tile the neuron into a small crossbar; implement a learning rule by firing write-pulses of light to potentiate/depress weights based on output error → a perceptron that trains itself with light.
3 · Tier 0 — build the working neuron (start here)
Parts (~$25, all hobby-grade)
| Part | Qty | Role |
|---|
| 5 mm LEDs — Red, Green, Blue | 3 | Inputs xᵢ (brightness = input value) |
| LDR / CdS photoresistor (GL5528) | 3 | Weights wᵢ (resistance = weight) |
| Color filter film (R/G/B gel, or colored cellophane) | 3 | Chromatic selectivity per synapse |
| LM358 op-amp | 1 | Summing / transimpedance amplifier (Σ) |
| LM393 comparator + 10k trimpot | 1 | Threshold → fire/no-fire |
| Resistors (1k–100k), breadboard, 5 V supply, output LED | — | Glue |
Build steps
- One synapse first: shine the Red LED through the red filter onto an LDR. The LDR + a fixed resistor form a divider; its current into the summing node ≈
x_R · w_R. Tune the LED brightness (input) and the partnering resistor (weight).
- Three colors: repeat for G and B. Mount each LED→filter→LDR in a short opaque tube (a straw painted black) so colors don't cross-talk. Optical isolation is the #1 thing that makes or breaks it.
- Sum: tie all three LDR outputs to the LM358 inverting input (virtual-ground summing amp, feedback resistor R_f). Output ≈
−R_f·(x_R w_R + x_G w_G + x_B w_B).
- Fire: feed the sum to the LM393; set the trimpot threshold. Output LED lights when Σ crosses it = the neuron "fires."
- "Train" by hand: adjust each weight (series resistor or filter density) until the neuron fires only for the color-pattern you want (e.g., "fire on lots of red + a little blue"). You've built and trained an optical chromatic perceptron.
4 · Tier 1 — the real optical memristor (ZnO persistent photoconductivity)
Why this is the real thing: a memristor must remember its conductance. ZnO's PPC gives you exactly that, set by light: blue/UV pulses = potentiation (weight ↑, persists), red/IR or gentle heat = depression/reset. Multi-level and analog = a real synaptic weight.
DIY ZnO synapse cell
- Electrodes: make interdigitated electrodes (IDE) — two comb-shaped contacts ~0.2–0.5 mm apart on glass or bare FR4/PCB (etch a comb, or use silver paint / a fine gold-leaf comb — ties to your silver-leaf work).
- ZnO film: bridge the gap with ZnO. Easiest DIY routes, low-tech first:
- Nanoparticle paint: disperse ZnO nanopowder in isopropanol + a drop of binder, drop-cast across the gap, dry, then anneal on a hotplate (~150–300 °C) to improve contact.
- Sol-gel: zinc acetate route, spin/drop-coat, anneal (cleaner films, more steps).
- Write/read: READ with a small DC bias (measure current = weight). WRITE potentiation with blue/UV LED pulses (385–405 nm); conductance climbs and persists. ERASE with heat or longer-wavelength light. Plot conductance vs. number of light pulses → your synaptic potentiation curve.
- Drop this cell in place of the Tier-0 LDR. Now the weight is written by light and remembered — a true optical memristor neuron.
Safety: 385–405 nm is near-UV — never look into it, wear UV-blocking glasses, keep exposure short. Anneal in ventilation. Treat the hotplate/UV as the two real hazards here.
Going chromatic (Tier 1→2)
One material rarely answers all colors. Two honest paths to chromatic weights:
- Filter path (simplest): identical broadband cells, each behind a R/G/B filter → each cell is one color-channel weight. Works immediately.
- Material path (deeper): different photoactive layers tuned to different bands — ZnO (UV/violet), a dye-sensitized or perovskite layer (visible), etc. Each color natively drives its own synapse. More research, more reward.
- Prism path (Tier 2): a glass prism or cut crystal splits a white input into its spectrum across a row of cells — one beam, many chromatic weights. This is your "cut-stone optical element," literally.
5 · Measuring & "thinking" you can show
- Weighted sum: hold weights fixed, sweep one color's brightness → output rises linearly until threshold = the neuron integrating.
- Chromatic selectivity: same total light, different colors → different firing = it "sees color," not just brightness.
- Memory (Tier 1): pulse blue, remove light, read hours later → weight persisted = non-volatile learning.
- Potentiation curve: conductance vs. #pulses — the classic synapse plot, from your own bench.
6 · Honest verdict
- Tier 0 — guaranteed to work. A real optical, chromatic, analog perceptron neuron. Best first build.
- Tier 1 (ZnO PPC) — real, published science; DIY-achievable but finicky (film quality, contact, and PPC retention drifts — weights slowly relax). Expect to iterate on the film. This is genuine optical-memristor behavior, not a gimmick.
- Tier 2–3 — the prism is easy and gorgeous; a self-training optical array is real but a project of months, not an afternoon.
- What's NOT here yet: a high-endurance, drift-free, many-level non-volatile optical memristor is still research-grade worldwide. Don't expect chip-like stability from a kitchen-bench ZnO cell — expect a working, teaching, exciting prototype that proves the principle.