Skip to content

Inexact Computing Learning & Research Portal

Welcome to the Inexact Computing Learning and Research Portal, an open-access living laboratory and interactive educational resource for approximate and inexact arithmetic computing.

graph TD
    A[Inexact Computing Hub] --> B[🎓 Fundamentals: History, Motivations & Math]
    A --> C[🌉 Bridge: Pixels & AI Resilience]
    A --> D[⚙️ Circuit Taxonomies: 7 Primitive Families]
    B --> E[Timeline, Grand Challenges & Error Bounds]
    C --> F[Perceptual Noise + Neural Nets]
    D --> G[Multipliers, Adders, CORDIC, Dividers, Transforms, Memory, AI Accelerators]

🎯 Objectives & Scope

Approximate computing trades tiny, bounded amounts of mathematical accuracy for massive reductions in power consumption, silicon area, and critical path delay in error-resilient applications: - Artificial Intelligence & Deep Learning (Conv2D, GEMM, Transformer attention, Systolic Arrays) - Computer Vision & Image Processing (Filtering, Edge detection, JPEG/HEVC DCT compression) - Signal Processing & Audio (FFT, FIR filters, CORDIC coordinate rotations) - Low-Power Memory Systems (Voltage-scaled SRAM caches, Refresh-relaxed DRAM)

Section Description
Glossary Plain-English definitions of core inexact computing concepts, acronyms, and terms.
Fundamentals The 1% Rule, History & Evolution (from Dennard scaling collapse to Palem's probabilistic computing), Motivations & Grand Challenges (AI climate crisis, verification nightmares, error containment), and mathematical error bounds.
Bridge to Real World Why human eyes forgive pixel noise (includes live Image Filter Sandbox) and why AI models thrive on inexact arithmetic.
Circuit Taxonomies Architectural guides and IEEE citations across 7 core hardware families: Multipliers, Adders, CORDIC, Dividers & Squarers, Transforms (DCT/FFT), Memory & Storage, and AI Accelerators.

🔬 Repository & Source Code