Netlist format

Seven bytes per gate, a flat node space, and a topological constraint. This is the entire encoding.

Byte layout

Node numbering

• 0 — constant 0
• 1 — constant 1
• 2 … 2+nIn-1 — circuit inputs, in order
• 2+nIn+i — output of gate i

Outputs are the last nOut nodes of the sequence. There is no separate output table; the ordering is the contract.

Evaluation

nodes = [0, 1] + inputs
for (op, a, b) in gates:
    x, y = nodes[a], nodes[b]
    nodes.append((1 - (x & y)) if op == 0 else x)

outputs = nodes[-nOut:]

NAND is 1 - (x AND y). LATCH passes x through, which is how a design can hold a value without introducing a new gate type.

A worked example

The first seven bytes of the ADD4 netlist are:

00 000002 000006
op=0 (NAND)   a=2   b=6

Node 2 is the first input, node 6 is the fifth input of a 9-input adder — the low bits of A and B. Its output becomes node 2 + 9 + 0 = 11.

Constraints

• Length must be a multiple of 7.
• Every referenced node must already exist: a, b < 2 + nIn + i for gate i. This keeps the list topological and forbids cycles.
• Node indices are 24-bit, which caps a single netlist at roughly 16.7 million nodes.

Size and cost

A 40-gate shifter is 280 bytes. A 2,118-gate multiplier is 14,826 bytes. The netlist is stored on chain as bytes, so tapeout gas scales with it, and so does the calldata of the tapeout transaction.

Why a flat list

It is the smallest encoding that can be decoded by a contract and re-executed by a browser without a compiler. There is no graph serialisation, no labels, no hierarchy — just gates in evaluation order.