Introduction to Systems Approach

Introduction to the Systems Approach

• The past 40 years have seen significant advances in silicon technology, leading to increased transistor density and performance.
• In 1966, Fairchild introduced a quad two-input NAND gate with approximately 10 transistors.
• In 2008, Intel's quad-core Itanium processor had 2 billion transistors.
• Transistor density has continuously improved, leading to a decrease in device cost.
• The aim of the book is to present an approach for computer system design that leverages advancements in transistor density.
• The approach builds upon previous studies in computer architecture and design, but it also delves into system architecture and design.
• The concept of systems engineering is introduced. It was developed in response to the increasing complexity of engineering problems in large-scale systems.
• A system-on-chip (SOC) is a complex architecture integrating various components, including processors, memories, and interconnects, tailored to a specific application.
• The Sony PlayStation 2's Emotion Engine is given as a simple example of an SOC architecture.
• This example contains a main processor (RISC style) and two vector processing units (VPUs) with four parallel processors each (SIMD style).
• The book will explore the components before returning to the issue of system architecture.

Components of the System

• System architecture defines the operational structure and the user's view of the system, including the functional specification and hardware implementation.
• The system architecture details system-level building blocks such as processors and memories, and the interconnection between them.

Processor Architectures

• Processors are characterized by their application and/or architecture.
• Processor examples include GPUs, DSPs, media processors, and network processors.
• Architectures such as SIMD, vector (VP), VLIW, and superscalar are categorized by their architecture or functionality.

Hardware and Software

• The fundamental decision in SOC design involves deciding which components are implemented in either hardware or software.
• Software implementations run on general-purpose processors which interpret instructions. This method is flexible but slower than dedicated hardware implementations.
• Hardware implementations, on the other hand, are faster and require less overhead of instruction fetching and decoding, but often lack programmability and more costly.
• SOC designs aim to combine the benefits of both for optimized performance.

Processor Architectures (continued)

• Coarse-Grained Reconfigurable Architecture (CGRA) includes logic blocks that process data.
• Structured ASICs allow for application customization before fabrication.
• Digital Signal Processors (DSPs) have fixed hardware architectures optimized for digital signal processing tasks.

Simple Sequential Processor

• Sequential processors execute instructions sequentially from the instruction stream.
• Instructions are executed sequentially to produce the desired result in a predefined order.
• Stages include fetching, decoding, generating the memory address, fetching operands, executing the operation, and writing back the result.

Pipelined Processor

• A pipelined processor overlaps the execution of multiple instructions. This process allows for multiple operations concurrently, thus increasing throughput.
• This process may include instruction fetch, decode, execution, etc, which are often performed simultaneously to maximize efficiency.
• A pipeline's stages are essential for completing one operation before the next operation begins.

Superscalar Processors

• Superscalar processors, unlike sequential processors, are designed to handle multiple instructions simultaneously.
• They use dynamic mechanisms to determine which operations can run in parallel, achieving a higher execution rate.

VLIW Processors

• VLIW processors rely on compilers to determine which operations can run in parallel, reducing the processor's complexity compared to Superscalar processors
• Instructions in VLIW processors contain multiple independent operations.
• They can achieve high performance from static scheduling, making them suitable for applications that are highly predictable.

Array Processors

• These processors contain multiple interconnected processing elements (PEs)
• Each PE has its own local memory and global memory space.
• Data and instructions are distributed regularly among PE's

Vector Processors

• Vector processors, similar to array processors, can operate on vectors, or sequences of data, as a single entity.
• They contain pipeline units that process vectors with high clock rates compared to scalar processors.
• These processors have special instructions and features for working efficiently on vectors for high performance.
• The handling of values between the vector register file and the main memory is performed concurrently to maximize throughput.
• These processors use techniques like "chaining," allowing computations to happen concurrently.

Multiprocessors

• Multiprocessors consist of multiple processors that work together and are interconnected to share resources.

• Data sharing and synchronization methods vary greatly depending on the architecture (distributed vs. shared memory). Multiprocessors are very complex

• Shared memory architectures are more complex because of memory consistency and cache coherency issues

SOC Design Approach

• Requirements and specifications are fundamental in system design.
• The process begins with understanding the design requirements.
• Different design stages are used: initial design and optimized design that iterate through steps.
• Iterative steps to refine and optimize performance