ISSCC 2026 Accelerator Survey

Interactive survey website for 43 accelerator/processor papers from ISSCC 2026 (Sessions 2, 10, 18, 30, 31).

Live site: https://devil-sx.github.io/isscc_accelerator/site/index.html

Features

• Overview table with sorting and filtering by session, process node, application, and keyword search
• Detail pages for each paper with:
  • Title annotation (说文解字) — color-coded breakdown of technical terms
  • Challenge-Idea cards — 3-4 paired problem/solution pairs per paper
  • Figure captions listing
  • Chip metrics — structured display of technology, area, power, efficiency, etc.
• Responsive layout for mobile and desktop

Project Structure

site/                  # SPA website (HTML/CSS/JS)
  index.html
  css/style.css
  js/app.js, overview.js, detail.js
data/                  # Structured data
  papers.json          # Main dataset (43 papers with metadata)
  {paper_id}/          # Per-paper directory
    figures.json       # Figure captions
    metrics.json       # Chip metrics
scripts/               # Data processing scripts
tests/                 # Playwright E2E tests

Note on Copyrighted Content

This public repository contains only derived metadata (titles, challenges, ideas, metrics, figure captions). Paper full text and figure images are excluded due to IEEE copyright. To use the full version locally:

1. Place source PDFs in pdfs/
2. Run python3 scripts/extract_all_figures.py to extract figures to images/
3. Run python3 scripts/restructure_data.py to generate text.md files
4. Run python3 scripts/update_papers_json.py to rebuild papers.json with image paths

The site auto-detects available content and gracefully handles missing images/text.

Public vs Private Mode

The site has two viewing modes, controlled by URL parameter — no code changes needed:

	Public (default)	Private (?private=1)
URL	site/index.html	site/index.html?private=1
Metadata & Abstract	Yes	Yes
Figure captions	Yes (with placeholder)	Yes (with images)
Figure images	No (not deployed)	Yes
Full paper text	No	Yes (three-mode reader)
Deployment	GitHub Pages	Local only

GitHub Pages serves the public version only — figure images (images/, images_web/) and paper full text (data/*/text.md) are gitignored due to IEEE copyright.

Private mode unlocks a three-mode reader per paper: paired figure+paragraph, full text, and image gallery. It requires local image and text files (see Copyrighted Content for setup).

Local Development

In VS Code: Ctrl+Shift+P → Tasks: Run Task → Serve (Local) — prints clickable URLs for both modes.

Or manually:

python3 -m http.server 8765
# Public:  http://localhost:8765/site/index.html
# Private: http://localhost:8765/site/index.html?private=1

Data Processing Scripts

# Extract figures from PDFs (requires PyMuPDF)
python3 scripts/extract_all_figures.py

# Restructure data directories and extract captions
python3 scripts/restructure_data.py

# Extract chip metrics from paper text
python3 scripts/extract_metrics.py

# Merge all data into papers.json
python3 scripts/update_papers_json.py