pub const FLOP_BIT_UM2: f64 = _; // 87.584999999999993f64Expand description
Silicon area per stored bit when an array is built from FLIP-FLOPS, in um^2.
Measured from the sky130 library’s own LEF, not estimated: dfxtp_1 is 20.019 um^2, plus the
access logic every addressed array needs — a mux2_1 read path at 11.261 and roughly a
nand2_1 of write decode at 3.754 — is 35.03 um^2 of CELL per bit. Standard cells do not fill
their area: these blocks place at FP_CORE_UTIL 30-40, so the die cost is that over ~0.40.
The comparison that matters: this is ~87 um^2/bit against a macro’s 17.4-20.5 um^2/bit, where the macro figure is its true LEF footprint with its own decode already inside. SRAM is about four times denser than flops. An earlier version of this constant was expressed in generic “cells” and made flops look CHEAPER than macros, which put a 512 kbit heap in flip-flops at 36 mm^2 and called it the smaller option.