Integrating neural architectures · low SWaP

Six architecture families.
One fabric.
An itemized energy bill.

Brain-inspired is an organization, not a biology. The 20-watt number does not come from spiking — it comes from locality, conditional compute, and recall over recompute. We have hands-on evidence, not a literature opinion: we built the direct spiking integration, and it taught us why that path doesn't pay yet.

The thesis

The spike is the accent, not the language

A neuron's electrical pulse is how a wet cell sends a signal down a wet wire. Reproducing it in silicon reproduces the mechanism that was never doing the work — and needs exotic hardware to do it. Three properties actually deliver the efficiency, and none of them is it spikes.

Locality

Compute and memory in the same place. Moving a weight from DRAM costs orders of magnitude more than the multiply it feeds. This is the largest term in any real power budget.

Conditional compute

About 2% of neurons are active at any instant. The cortex is dark almost everywhere, almost always. You don't wake the visual system to smell coffee.

Recall over recompute

Most cognition is retrieval of consolidated belief, not fresh inference. You don't re-derive that fire is hot. Recall is orders of magnitude cheaper.

This isn't theory. It is the fourth iteration.
We have been building brain-inspired cognitive architectures for over a year — memory held separately from the talker, a cheap layer gating an expensive one, facts written and retrieved rather than baked into weights. Successive systems tested those ideas on different substrates, including a direct implementation on a spiking neuroscience engine. The principles held every time. What did not is any approach that pays for biological fidelity up front and waits for the efficiency to arrive. WICK is where those principles meet engines we own outright.
The fabric

Cheapest first, and any stage can end the turn

The topic's own complaint is that current AI is monolithic. Its ask is a floor plan: specialized areas, arranged in a hierarchy, work routed to whichever the task needs. The transformer — which in a conventional system is the whole architecture — is 98.9% of this fabric's dense cost and the last thing consulted.

SALIENCEspiking 8K MACs
SEQUENCErecurrent 180K
RECALLVSA / hyperdimensional 30K
ONTOLOGYsymbolic graph 12K
PERCEPTIONconvolutional 4.2M
ASSOCIATIONtransformer 29.8M

One ear for six species

Interconnecting modules at the output end is easy. At the input end is the hard half — solved by a single deterministic character-to-waveform transducer, so a convolutional region and an algebraic region look at the same signal. It's also what makes event-driven processing possible: you can't difference what was never continuous.

Knowing is separate from saying

Recall holds what is true, in rows a human can read, edit and delete. The transformer holds how to say it and is never the source of a fact. That's what makes skipping it safe rather than degrading — and why the honest answer is "I do not have that" instead of a confabulated near-match.

Train the talker, and only the talker.
One region is trained, for fluency only — how to say a thing, never what is true. Facts are written as rows; the hierarchy is declared; salience and sequence hold no content at all. A model that never has to carry the world's facts doesn't need the parameters that carrying them requires, which is why 29.8M trained from random weights on one desktop is enough. And a wrong fact is one editable row, not a weight nobody can find — at the edge, that's the difference between a system you can update and one you have to replace.

The one thing we haven't trained yet

Separating knowing from saying creates a skill neither half has: speaking from a retrieved structure. Hand a model a row and expect fluent, faithful speech out the other side — that's an assumption, and it's the seam where this architecture tears first. An untrained talker will ignore the chip and answer from its weights, paraphrase it into something subtly untrue, or read it out like a database.

The corpus is a by-product of a system we already run. EMBER carves chips from conversation on the way in, so for any real exchange we can see exactly which chips were written and exactly what was said next. The pair isn't constructed, it's recorded — no labeling pass, no synthetic generation, no large model in the loop teaching a small one its confident wrongness.

ConstructionWhat the model must learn
full chip set → the real reply faithful speech from structure
hold back a chip → the reply that used it must no longer assert it
empty chip set → any question "I do not have that."
low-confidence chip → the reply hedge, do not assert
inherited chip → the reply state it, and that it was inferred

Ablating a chip and requiring the model to drop the claim it supported is the direct cure for the worst failure mode: answering from weights while pretending to answer from memory. A model trained this way can't smooth over a missing row — it saw thousands of examples of the row being missing.

Requirement (2) — dynamic allocation

A brain that keeps its own health, and logs it

Lifted from production training code where it decides per-layer precision during real runs, re-aimed at per-region allocation. Every region carries a health bar: low demand damages it, genuine work regenerates it.

BREAK

Health hits zero, the region changes precision regime permanently. No flapping back — oscillating pays the switching cost forever and gets neither regime's benefit.

HARDEN

A region that absorbs pressure without breaking gets tougher, scaled by how close the hit came. Surviving a scare at 4 bits teaches it to shrug off what would have staggered it before.

CONDUCT

One component sees the whole workload rather than one region, declares its phase, and sets the rules the breakers judge under. Phase changes need consecutive agreeing readings.

  phase=ACTIVE step=10
  salience      4b health 10.0 hits 0   hardens 0
  sequence      4b health 10.0 hits 0   hardens 0
  recall        4b health 10.0 hits 0   hardens 0
  ontology      4b health 10.0 hits 0   hardens 0
  perception    4b health  5.0 hits 5   hardens 1
  association   4b health 10.0 hits 0   hardens 0
  log: step 4: phase COLD -> ACTIVE (novelty 0.25)

An efficiency claim that arrives as a single benchmark number cannot be interrogated. A running health ledger can. BREAK perception at step 12 is a sentence a reviewer can read.

Requirement (3) — counted, not asserted

The same regions, the same work, both levers off

Every region reports the multiply-accumulates it actually performed and the precision it performed them at. The ledger charges MACs × bits — deliberately conservative, since memory traffic scales linearly with bits moved while multiply energy scales worse than linearly. If it wins under pessimistic accounting, it wins.

75.0%
avoided on novel input, every region woken. Precision alone.
96.8%
avoided on familiar input, transformer never woken. Precision plus conditional compute.
8
MACs of 8,000 in the salience region when the input hasn't changed. Event-driven, no spiking silicon.
  region                 MACs      bits    units    why
  perception         4,200,000      4 16,800,000    feature extraction
  sequence             180,000      4    720,000    temporal state
  ontology              36,000      4    144,000    cheetah -> mammal -> animal
  recall                12,288      4     49,152    bind/unbind/cleanup
  salience                   8      4         32    delta 0.00
  association               --     --          0    SKIPPED: demand 0.05 below floor
  TOTAL              4,428,296             17,713,184

Ontology pays for itself. Five facts about a cheetah are returned when only one was ever stored — the rest inherited through cheetah → mammal → animal for 12K MACs. Every edge walked is a 29.8M-MAC wake that doesn't happen. It toggles off; it defaults on.

What is not proven

Stated plainly, because hidden unknowns aren't fundable

ClaimStatus
Fabric, governor, ledger, transducer, ontology complete · 21 tests
FP4 / nv4 substrate, trained from scratch real · running
Architectural reduction ratios above measured in-repo
Conv & transformer region backends cost models & interfaces; capability in sibling repos
INT2 at 2.32 effective bits projected arithmetic — NOT measured
Absolute power in watts needs a board — Phase II

On INT2. FP4 is viable because a higher tier holds its signposts. The same move goes one rung down: keep a ~16% high-magnitude tail at FP4, crush the remaining 84% to 2 bits, for 0.16×4 + 0.84×2 = 2.32 effective bits and a further 42% cut. A first attempt has not demonstrated it. If an FP4 anchor proves too thin, FP8 signposts give 2.96 bits at a known cost. Both rungs are stated as arithmetic in the source so the projection can be checked. Neither is claimed as a result.