A serial transfer moves an n-bit word one bit per clock edge, using just one wire. Connect register A's serial output to register B's serial input and shift both together. After n edges, A's old contents are in B.
Compare a parallel transfer (B ← A in register transfer notation), which uses n wires and a single edge. The trade-off:
- Serial: 1 wire, n edges. Cheap on wiring, slow.
- Parallel: n wires, 1 edge. Fast, but needs many connections.
That is why long-distance and chip-to-chip links (USB, SPI, UART, PCIe lanes) are serial, while transfers inside a CPU are parallel.
Tracing rule: at each edge, B's first stage takes A's old last bit, by the usual edge rule. Whatever A's own serial input carries fills A behind the departing bits. If A's serial output is also wired back to A's serial input, A keeps its value (it rotates back to the start after n edges) while B receives a copy.
Order matters: in a shift-right setup the bits travel LSB first. A receiver must shift in the same direction to rebuild the word correctly.