010010010
Σ MATCH ≥ 9OFFICIAL MACHINE 02 / NAND-NATIVE CNN
BC-01 / ORIENTATION ENGINE
Four 3×3 binary kernels scan an 8×8 pixel field. Every receptive field passes through XNOR, POPCOUNT, and threshold logic before full NAND/LATCH synthesis.
LIVE INFERENCE8×8 → 3×3×4 → 6×6×4
PIXEL FIELD8×8 / 64 BIT
0001000000010000000100000001000000010000000100000001000000010000
SHARED WEIGHTS4 KERNELS / 36 WINDOWS
000111000
Σ MATCH ≥ 9100010001
Σ MATCH ≥ 9001010100
Σ MATCH ≥ 9WINDOWR0 C0
WEIGHT SHARING / SAME KERNEL · 36 POSITIONSFEATURE MAPS4 × 6×6
GLOBAL OR POOL
V1H0D0A0
CLASS VRECEPTIVE FIELDS144
XNOR MATCHES1,296
NAND—
LATCH—
NETLIST SHA-256—
ONE MODEL. ONE MACHINE. ANY EXECUTOR.
Model machine fabrication and verification
A weight is a wire. A neuron is a circuit. A model is a machine. Once sealed, anyone can run it, anyone can verify it, and nobody can alter it.
SIGNAL LIVE16→8→4
CLASS OWEIGHT FIELDMOVE TO ISOLATE PATH
MODELNF-01 / GENESIS
16 → 8 → 4- Fixed weights
- 160
01Fixed weightsW ∈ {0,1}
→02XNOR + POPCOUNTBinary neuron
→03NAND / LATCHGate netlist
→04SHA-256Verifiable circuit
MODEL GRAPH
Model graph and weight matrix
bnn-model@1INPUT VECTOR011010011001011016 BIT
LAYER 00GLYPH FEATURE BANK
16 → 8N0
0110011001100110
Σ8 ≥ 140N1
1110011001100111
Σ6 ≥ 140N2
1000100010001111
Σ9 ≥ 140N3
1001100010000111
Σ9 ≥ 140N4
1001011001101001
Σ0 ≥ 140N5
1101011001101011
Σ2 ≥ 140N6
0110100110010110
Σ16 ≥ 141N7
0110110110110110
Σ14 ≥ 141LAYER 01GLYPH CLASSIFIER
8 → 4N0
11000000
Σ4 ≥ 70N1
00110000
Σ4 ≥ 70N2
00001100
Σ4 ≥ 70N3
00000011
Σ8 ≥ 71LIVE INFERENCE
Layer activation lab
2 CLOCKSPIXEL INPUT / 4×4
OFFICIAL TEST VECTORSINFERENCE CLASS O
t0IN
0110100110010110t1L0
00000011
8 / 6 / 9 / 9 / 0 / 2 / 16 / 14t2L1
0001
4 / 4 / 4 / 8Inference output
I0L0X0O1
CLASS OGATE COMPILER
BNN → NAND synthesis pipeline
IDLE01BNN MODEL16 → 8 → 4
02LAYER ELABORATION160 fixed weights
03LATCH PIPELINE2 clock stages
04NAND TECHMAPXNOR / ADDER / LATCH
- XNOR
- —
- HA / FA
- —
- NAND
- —
- LATCH
- —