Computer architecture borrowed its name from building design, and its logic too
In 1959 an IBM researcher writing about the Stretch supercomputer decided that machine organization sounded too dull, and called his subject system architecture instead. The label stuck because it fit: like a building architect, a computer architect works out what users need and designs to meet it within limits of money, power and physics.
The discipline has deep roots. Charles Babbage and Ada Lovelace described the structure of the analytical engine in their letters, and in 1936 Konrad Zuse, while building his Z1, noted in patent filings that a machine's instructions could sit in the same memory as its data, the stored-program idea. In 1945 John von Neumann's draft report on the EDVAC and Alan Turing's more detailed proposal for the Automatic Computing Engine, which cited it, set out how an electronic computer's logic might be arranged.
Lyle Johnson coined the term, and his colleague Frederick Brooks defined it, calling it the art of determining a structure's users' needs and meeting them economically. Brooks then helped build IBM's System/360, where architecture came to mean what a programmer needs to know. Machines descended from that line, today's IBM Z, still run compatible software.
Architects now split the field into layers. The instruction set is the contract between software and hardware: the numbered commands a processor understands, its registers, data types and ways of addressing memory. Compilers translate programs into those commands, and assemblers turn short human-readable mnemonics into binary. Beneath that sits the microarchitecture, the way a particular chip carries out the instruction set; cache size, for instance, is invisible to it. Microcode adds flexibility, letting a redesigned chip present the old instruction set so existing software keeps working.
Every choice is a trade. Richer instructions pack more work into fewer bytes of code but take longer to decode, cost more hardware and interact in surprising ways. Designers balance speed against power, cost, latency and reliability, and power budgets now often dominate. Early machines were designed on paper and built directly; since the 1990s new designs usually run first inside simulators or reprogrammable FPGA chips before anyone commits to silicon, and test chips may still need several redesigns.
Source: Computer architecture
