Netlist format
A TapeOut circuit's netlist is a list of two-input NAND gates: pairs says which two wires each gate connects to, taps says which wires the outputs come from.
Fields
| Field | Type | Meaning |
|---|---|---|
pairs | [[a, b], …] | NAND gates in order. pairs[i] = [a, b] means this gate's output = NAND(wire a, wire b). Required |
taps | [int, …] | Which wire output o comes from. Required |
nIn | int | Number of inputs |
nOut | int | Number of outputs, equal to len(taps) |
g | int | Number of gates, equal to len(pairs) |
d | int | Depth: the largest depth among all output wires |
C | int | Cost g · d³ |
srcId or id | int | Problem number. Different files use different key names for the same thing |
last_outs | bool | Whether the outputs are exactly the last nOut gates |
The on-chain format adds:
| Field | Meaning |
|---|---|
enc | "chain" |
n_base | 2 + nIn: the constants ZERO and ONE plus the inputs |
nl_hex | The on-chain netlist bytes in hexadecimal, starting with 0x |
nl_len | Number of bytes nl_hex decodes to |
Wire numbers
There are two numberings; they differ in the constants.
Inputs from 0 (common in design tool output):
| Wire | What it is | Depth |
|---|---|---|
0 … nIn−1 | Inputs in0 … in{nIn−1} | 0 |
nIn + i | Gate i | 1 + max(depth a, depth b) |
On-chain format (constants take 0 and 1):
| Wire | What it is | Depth |
|---|---|---|
0 | Constant ZERO, not a gate | 0 |
1 | Constant ONE, not a gate | 0 |
2 … nIn+1 | Inputs | 0 |
n_base + i | Gate i, counted in g | 1 + max(depth a, depth b) |
Gates
val[w] = 1 − (val[a] & val[b])- There is only one kind of gate: the two-input NAND.
NAND(x, x) = NOT(x). That is how inversion works; there is no separate polarity field.- A gate's two inputs may only refer to wires that already exist: inputs, constants or earlier gates.
- The order of the two inputs doesn't change the result, and no order is enforced.
Depth and cost
- Inputs and constants have depth 0; a gate's depth = 1 + the larger depth of its two inputs.
d= the largest depth among all output wires.C = g · d³.gdoesn't count the constants ZERO and ONE.
Outputs must be the last gates
The on-chain format requires taps to be exactly the last nOut gates:
base = nIn (inputs from 0) or n_base (on-chain format)
taps == [base + g − nOut, …, base + g − 1], and 0 < nOut ≤ gIf not, convert first: copy the output gates to the end, then delete gates nobody uses. Conversion may add gates or depth, so the cost is computed from g and d after conversion.
On-chain netlist bytes (nl_hex)
A pure NAND netlist uses 7 bytes per gate:
| Offset | Length | Content |
|---|---|---|
| 0 | 1 | opcode, 0 = NAND2 |
| 1 | 3 | Wire number of input a, big-endian |
| 4 | 3 | Wire number of input b, big-endian |
Total length = g × 7. Wire numbers are in the on-chain format.
Truth table bit order
in0is the lowest bit of the input combination number.- In the hexadecimal of the official vectors,
in_kis bitk % 8of bytek // 8.
Example
One circuit with 6 inputs and 3 outputs, in both formats:
| Item | Inputs from 0 | On-chain format |
|---|---|---|
nIn, nOut | 6, 3 | 6, 3 |
g, d | 47, 8 | 47, 8 |
C | 24,064 = 47 × 8³ | Same as left |
n_base | — | 8 |
| First three gates | [0,0], [2,2], [6,7] | [2,2], [4,4], [8,9] |
taps | [50, 51, 52] = starting at 6 + 47 − 3 | [52, 53, 54] = starting at 8 + 47 − 3 |
nl_len | — | 329 = 47 × 7 |
Start of nl_hex | — | 0x00 000002 000002 … (the first gate, NAND(2, 2)) |
Invalid netlists
len(taps) ≠ nOutg ≠ len(pairs)C ≠ g · d³- A gate refers to a wire that only comes later (a loop, or undefined)
Not decided yet
- The full table of on-chain opcodes besides
0(LATCH and others for sequential circuits). - An explicit version field
schema_versionand a wire-numbering fieldwire_base: only proposals, not used yet.