Computer architecture

Computer architecture is the fundamental structure of a computer system. It is the science of organizing component parts to create a working machine. Designers use architecture to balance several competing factors. These include performance, efficiency, cost, and reliability. A well-designed architecture ensures that a computer can handle tasks effectively. It must also work within specific technological and economic constraints.

There are three primary subcategories within this discipline. The first is the Instruction Set Architecture, or ISA. The ISA acts as the interface between a computer's software and its hardware. It defines the machine code that a processor reads and acts upon. It also sets the word size, memory address modes, and processor registers. The second subcategory is microarchitecture, sometimes called computer organization. This describes how a specific processor implements the ISA. For example, the size of a CPU cache is a microarchitecture decision. The third subcategory is systems design. This includes all hardware components beyond the CPU, such as multiprocessing or direct memory access.

The mechanism of the ISA is vital for how computers function. Most people write programs in high-level languages like Java or C++. However, a processor cannot understand these languages directly. Instead, software tools called compilers translate these languages into numerical instructions. These instructions are usually encoded as binary numbers. An assembler is another tool that translates human-readable forms of the ISA into computer-readable forms. The ISA provides the rules for how these instructions interact with memory and data types. This creates a bridge that allows software to control physical hardware.

History shows that computer architecture has evolved from paper to complex simulations. The first documented architecture appeared in letters between Charles Babbage and Ada Lovelace regarding the analytical engine. In 1936, Konrad Zuse described the stored-program concept. This idea allows machine instructions to be stored in the same memory used for data. In 1945, John von Neumann published a report describing the organization of logical elements. That same year, Alan Turing proposed a detailed design for an electronic calculator. The specific term "architecture" gained popularity at IBM in 1959. Researchers Lyle R. Johnson and Frederick P. Brooks, Jr. used it to describe system needs and hardware parameters.

Modern implementation follows a rigorous, multi-step process. First, engineers create designs on paper or in simulators. They might use a Field-Programmable Gate-Array, or FPGA, to build a physical prototype. This allows them to test and tweak the design before committing to final hardware. Once the design is ready, engineers move to implementation. This includes logic implementation, which designs circuits at a logic-gate level. Next is circuit implementation, which involves transistor-level designs of components like ALUs or caches. Finally, physical implementation draws the actual circuits and places them on a chip floor plan. The entire process concludes with design validation to ensure the computer works correctly.

Designers must make difficult trade-offs during the ISA design phase. A complex instruction set can make programs more space-efficient. For instance, a single instruction might encode a high-level task like a loop. However, complex instructions take longer for the processor to decode. This can increase the cost of the hardware and decrease speed. High complexity can also lead to unreliability if instructions interact in unexpected ways. Therefore, a good ISA must find a balance between programmer convenience, code size, cost, and speed. Designers use emulators to measure these factors during the design process.

Beyond the three main subcategories, other specialized technologies exist. In 2002, it was estimated that technologies like macroarchitecture and microcode accounted for 1% of computer architecture. Macroarchitecture involves layers that are more abstract than microarchitecture. Microcode is a type of software that acts as a wrapper around the hardware. It translates instructions to allow a new chip to run old software without changes. Another specialized area is pin architecture. This defines the hardware functions a microprocessor must provide to a platform. These functions are flexible because external hardware can adapt to new encodings or messages. This ensures that different components remain compatible with each other.