1:8 demultiplexer threshold circuit, magnitude 52

README.md

@@ -0,0 +1,65 @@
+---
+license: mit
+tags:
+- pytorch
+- safetensors
+- threshold-logic
+- neuromorphic
+---
+
+# threshold-demux8
+
+1:8 demultiplexer. Routes data input to one of 8 outputs based on 3-bit select.
+
+## Function
+
+DEMUX8(d, s2, s1, s0) -> [y0, y1, ..., y7]
+
+yi = d AND (s == i), where s = 4*s2 + 2*s1 + s0
+
+## Architecture
+
+Single layer with 8 neurons. Each output yi fires when d=1 AND select matches i.
+
+| Output | Weights [d, s2, s1, s0] | Bias |
+|--------|-------------------------|------|
+| y0 | [1, -1, -1, -1] | -1 |
+| y1 | [1, -1, -1, +1] | -2 |
+| y2 | [1, -1, +1, -1] | -2 |
+| y3 | [1, -1, +1, +1] | -3 |
+| y4 | [1, +1, -1, -1] | -2 |
+| y5 | [1, +1, -1, +1] | -3 |
+| y6 | [1, +1, +1, -1] | -3 |
+| y7 | [1, +1, +1, +1] | -4 |
+
+## Parameters
+
+| | |
+|---|---|
+| Inputs | 4 (1 data + 3 select) |
+| Outputs | 8 |
+| Neurons | 8 |
+| Layers | 1 |
+| Parameters | 40 |
+| Magnitude | 52 |
+
+## Usage
+
+```python
+from safetensors.torch import load_file
+import torch
+
+w = load_file('model.safetensors')
+
+def demux8(d, s2, s1, s0):
+    inp = torch.tensor([float(d), float(s2), float(s1), float(s0)])
+    return [int((inp * w[f'y{i}.weight']).sum() + w[f'y{i}.bias'] >= 0)
+            for i in range(8)]
+
+# Route d=1 to output 5 (s=101)
+print(demux8(1, 1, 0, 1))  # [0, 0, 0, 0, 0, 1, 0, 0]
+```
+
+## License
+
+MIT

config.json

@@ -0,0 +1,9 @@
+{
+  "name": "threshold-demux8",
+  "description": "1:8 demultiplexer as threshold circuit",
+  "inputs": 4,
+  "outputs": 8,
+  "neurons": 8,
+  "layers": 1,
+  "parameters": 40
+}

create_safetensors.py

@@ -0,0 +1,51 @@
+import torch
+from safetensors.torch import save_file
+
+weights = {}
+
+# Input order: d, s2, s1, s0 (4 inputs)
+# Output: y0-y7
+# yi = d AND (s == i)
+
+for i in range(8):
+    w = [1.0]  # d weight
+    s2_bit = (i >> 2) & 1
+    s1_bit = (i >> 1) & 1
+    s0_bit = i & 1
+    w.append(1.0 if s2_bit else -1.0)  # s2
+    w.append(1.0 if s1_bit else -1.0)  # s1
+    w.append(1.0 if s0_bit else -1.0)  # s0
+    bias = -(1 + bin(i).count('1'))
+
+    weights[f'y{i}.weight'] = torch.tensor([w], dtype=torch.float32)
+    weights[f'y{i}.bias'] = torch.tensor([bias], dtype=torch.float32)
+
+save_file(weights, 'model.safetensors')
+
+# Verify
+def demux8(d, s2, s1, s0):
+    inp = torch.tensor([float(d), float(s2), float(s1), float(s0)])
+    outputs = []
+    for i in range(8):
+        y = int((inp * weights[f'y{i}.weight']).sum() + weights[f'y{i}.bias'] >= 0)
+        outputs.append(y)
+    return outputs
+
+print("Verifying DEMUX8...")
+errors = 0
+for d in [0, 1]:
+    for s in range(8):
+        s2, s1, s0 = (s >> 2) & 1, (s >> 1) & 1, s & 1
+        result = demux8(d, s2, s1, s0)
+        expected = [1 if (d == 1 and i == s) else 0 for i in range(8)]
+        if result != expected:
+            errors += 1
+            print(f"ERROR: d={d}, s={s}, got {result}, expected {expected}")
+
+if errors == 0:
+    print("All 16 test cases passed!")
+else:
+    print(f"FAILED: {errors} errors")
+
+mag = sum(t.abs().sum().item() for t in weights.values())
+print(f"Magnitude: {mag:.0f}")

model.py

@@ -0,0 +1,22 @@
+import torch
+from safetensors.torch import load_file
+
+def load_model(path='model.safetensors'):
+    return load_file(path)
+
+def demux8(d, s2, s1, s0, weights):
+    """1:8 Demultiplexer: routes d to output y[s] where s = 4*s2 + 2*s1 + s0"""
+    inp = torch.tensor([float(d), float(s2), float(s1), float(s0)])
+    outputs = []
+    for i in range(8):
+        y = int((inp * weights[f'y{i}.weight']).sum() + weights[f'y{i}.bias'] >= 0)
+        outputs.append(y)
+    return outputs
+
+if __name__ == '__main__':
+    w = load_model()
+    print('DEMUX8 verification:')
+    for s in range(8):
+        s2, s1, s0 = (s >> 2) & 1, (s >> 1) & 1, s & 1
+        result = demux8(1, s2, s1, s0, w)
+        print(f'  d=1, s={s} ({s2}{s1}{s0}) -> {"".join(map(str, result))}')