Intel released the 8087 floating‑point coprocessor in 1980, boosting math speed up to 100×. Its core was a 69‑bit adder that handled arithmetic, roots, and transcendental functions such as tangent and logarithm. The adder lived inside a nanomachine of registers, shifters, and control logic, as the patent shows on a 5 mm×6 mm die with 40 external pins.

Inside the die, the top layer hosts a Bus Interface Unit that ties the chip to an 8086 host and memory. A large microcode ROM defines the 8087’s instruction set. The datapath splits into exponent and fraction sections; the fraction path centers on the adder, which must propagate carries across 69 bits quickly in less than a microsecond.

Achieving speed required breaking the 69‑bit addition into 4‑bit blocks. Each block uses a Manchester carry chain, a technique from the 1959 Atlas computer that pre‑sets switches to let carries flow at wire speed. The chain handles generate, propagate, and delete cases in parallel, reducing ripple delay by a factor of four in a 1 ns.

To keep the chain fast, the 8087 pre‑charges carry lines to 5 V at each addition start, then pulls them low if needed. A carry‑skip circuit detects when all 4 bits propagate, allowing the carry to jump to the next block. This design lets the 8087 perform complex math with unprecedented precision for its time in.