Computer architecture

Computer architecture
Amila Ruwan 20 · CC BY-SA 4.0 · source
Name	Computer architecture

Computer architecture Computer architecture is the structured design and organization of computing systems, encompassing the arrangement of processors, memory, interconnects, and peripherals to execute programs efficiently. It links the conceptual models of computation championed by pioneers like Alan Turing, John von Neumann, Claude Shannon, Donald Knuth, and Grace Hopper with practical implementations from organizations such as Intel Corporation, Advanced Micro Devices, IBM, ARM Holdings, and NVIDIA Corporation. The field draws on advances from institutions including Massachusetts Institute of Technology, Stanford University, University of California, Berkeley, Bell Labs, and Los Alamos National Laboratory and informs products used in markets led by Microsoft Corporation, Apple Inc., Google LLC, Amazon.com, Inc., and Samsung Electronics.

Overview and history

Early designs trace to machines built by Charles Babbage and theoretical models by Alonzo Church and Turing. The Manchester Baby and projects at ENIAC set foundations that influenced the von Neumann architecture adopted in early systems from IBM, UNIVAC, and Digital Equipment Corporation. Mid-20th-century innovations by Maurice Wilkes, John Backus, Edsger Dijkstra, and John Cocke enabled compiler optimizations, pipelining, and RISC philosophies embodied by Berkeley RISC and Stanford MIPS efforts. The microprocessor revolution led by Intel 4004 and later Intel 8086 spawned personal computing ecosystems around Commodore, Apple Computer, Microsoft, and IBM PC standards. The rise of parallelism involved work from Seymour Cray, Gene Amdahl, Gordon Moore, and institutions like Los Alamos National Laboratory and Oak Ridge National Laboratory, culminating in supercomputers such as Cray-1 and exascale initiatives by Oak Ridge and Argonne National Laboratory.

Fundamental concepts

Architectural abstractions include the instruction set, datapath, control unit, register file, cache hierarchy, and system buses, influenced by theories from John von Neumann, Claude Shannon, Alan Turing, Noam Chomsky (in formal languages), and Donald Knuth. Key principles—pipelining, superscalar execution, speculative execution, out-of-order execution, and branch prediction—were advanced by researchers at IBM Research, Bell Labs, HP Labs, and Intel Corporation. The design trade-offs between complexity and performance are governed by models and laws such as Amdahl's law, Gustafson's law, and Moore's law, with impacts studied in programs at MIT CSAIL and UC Berkeley RISE Lab.

Processor design and instruction set architecture

Processor design encompasses datapath organization, microarchitecture, pipeline depth, execution units, and control logic developed in projects like ARM Cortex, Intel Core, AMD Ryzen, IBM POWER, and NVIDIA Tegra. Instruction set architecture (ISA) families—such as x86, ARM, RISC-V, MIPS, and SPARC—define programmer-visible operations and calling conventions used in systems from Dell Technologies, HP Inc., Lenovo, and ASUS. Microarchitectural techniques from John Cocke and groups at IBM include register renaming, Tomasulo's algorithm, and the reorder buffer. Compiler-targeting strategies evolved in work by Gerwin Klein, Frances Allen, and Ken Thompson enabling toolchains like GCC, LLVM, and Microsoft Visual C++ to exploit ISA features.

Memory hierarchy and storage

Memory systems integrate register files, multiple cache levels, main memory, and persistent storage using technologies from Micron Technology, SK Hynix, Samsung Electronics, and Western Digital. Cache coherence protocols (MESI, MOESI) and NUMA architectures emerged from research at HP Labs, Intel Research, and Cray Research. Virtual memory concepts implemented in systems by DEC, Sun Microsystems, and IBM use page tables, TLBs, and demand paging; related algorithms were analyzed by Peter Denning and Tom Leighton. Secondary storage innovations include solid-state drives driven by Samsung, Intel, and Seagate Technology and distributed storage models used by Google File System, Hadoop, and Ceph.

Input/output and interconnects

I/O systems and interconnects cover buses, controllers, DMA, interrupts, and network fabrics found in product lines from Intel, Broadcom, Qualcomm, and NVIDIA. Interconnect topologies—ring, mesh, fat-tree, torus—are applied in systems developed by Cray, IBM, and Cisco Systems. High-performance fabrics such as InfiniBand and Ethernet enhancements are used in data centers of Google, Amazon Web Services, and Microsoft Azure. Peripheral standards like PCI Express, USB, SATA, and Thunderbolt link devices from Apple Inc., Dell, and HP to host controllers; controllers and switches implement flow control and congestion management techniques advanced at Stanford and MIT.

Performance evaluation and benchmarking

Performance analysis uses metrics and tools influenced by standards from SPEC and workloads developed by PARSEC, TPC, and benchmarks like LINPACK and STREAM. Profiling and tracing tools from Intel VTune, perf, Valgrind, and GProf help researchers at Berkeley and MIT evaluate designs. Evaluation methodologies leverage queuing theory, simulation frameworks such as GEM5, and cycle-accurate models developed in academia and industry including ARM Research and IBM Research. Notable studies on multicore scaling cite Amdahl's law and Gustafson's law and use platforms from NVIDIA and AMD to measure throughput, latency, energy efficiency, and reliability.

Emerging trends and future directions

Current directions span heterogeneous computing with accelerators from NVIDIA, AMD, and Intel; open ISAs like RISC-V promoted by SiFive; photonic interconnect research at MIT and Caltech; near-memory and in-memory computing initiatives by IBM Research and Samsung; and quantum-inspired co-design in collaborations involving Google Quantum AI, IBM Quantum, Microsoft Quantum, and Rigetti Computing. Edge and IoT deployments led by ARM Holdings, Qualcomm, and NVIDIA Jetson focus on energy efficiency and security techniques from DARPA programs and standards by IEEE. Research into neuromorphic systems by Intel Labs' Loihi project and IBM TrueNorth, and into chiplet architectures advanced by AMD and the U.S. Department of Defense signal continuing evolution in scalability, verification, and sustainability.

Category:Computer science