Computer architecture is the design and organization of computing systems, encompassing how hardware components work together to execute instructions and process data. It represents the conceptual blueprint and structural framework that defines a computer system’s functionality, performance, and capabilities.
Fundamentals
Von Neumann Architecture
- Single memory space for both instructions and data
- Sequential instruction execution
- Characterized by the “von Neumann bottleneck” between processor and memory
- Still forms the foundation of most modern computing systems
Harvard Architecture
- Physically separate storage and signal pathways for instructions and data
- Allows simultaneous access to both instruction and data memory
- Common in DSPs and microcontrollers
- Reduces the memory bottleneck for certain applications
Instruction Set Architecture
- ISA: The interface between hardware and software, defining the set of instructions a processor can execute, addressing modes, registers, and data type.
Instruction Cycle (Fetch-Decode-Execute)
CPU Design
- ALU (Arithmetic Logic Unit)
- Control Unit
- Registers
- Pipelining & Hazards
Memory Hierarchy
- Registers
- Cache (L1, L2, L3)
- RAM
- Virtual Memory & Paging
Memory-Centric Architectures
- Non-von Neumann approaches
- Neuromorphic computing
- In-memory computing
- Near-data processing
I/O System
- System Buses (Data, Address, Control)
- I/O Interfaces
- Interrupts & DMA
Performance
- Clock Speed & CPI
- Benchmarks & MIPS/FLOPS
Parallel Processing Architectures
- SIMD (Single Instruction, Multiple Data): Same operation performed on multiple data points simultaneously
- MIMD (Multiple Instruction, Multiple Data): Multiple processors executing different instructions on different data
- Vector Processing: Specialized for operations on data arrays
- GPU Architecture: Optimized for graphics and highly parallel computation
Modern Architectures
RISC (Reduced Instruction Set Computer)
- Simple, highly-optimized instruction set
- Fixed instruction length
- Large number of general-purpose registers
- Load-store architecture
- Examples: ARM, RISC-V, early MIPS
CISC (Complex Instruction Set Computer)
- Rich instruction set with many specialized instructions
- Variable instruction length
- Fewer registers
- Memory-to-memory operations
- Example: x86 architecture
Superscalar & Out-of-Order Execution
Specialized Architectures
- FPGA-based: Reconfigurable hardware
- ASIC: Application-specific integrated circuits
- Quantum Computing: Based on quantum mechanical principles
- AI Accelerators: Specialized for machine learning workloads (TPUs, NPUs)