Datasheet Intelligence

Core Workflow

1. Identify the component and datasheet structure.

   • Extract cover page: part number, manufacturer, datasheet revision date.
   • Identify major sections: Absolute Maximum Ratings, Electrical Characteristics, Typical Performance Characteristics (graphs), Application Circuits.
   • Load references/datasheet-structure.md for section layout by component type.

2. Extract parameter tables.

   • Run scripts/extract_datasheet_tables.py --file datasheet.pdf.
   • The script uses pdfplumber with tight table detection tuned for electrical characteristic tables (header recognition: "Parameter", "Symbol", "Min", "Typ", "Max", "Unit", "Condition").
   • Output: one CSV per table, plus a Markdown summary with table names inferred from surrounding text.
   • Load references/datasheet-structure.md for parameter naming conventions.

3. Identify curves to digitize.

   • First extract images with pdf-reader skill: pdf_extract_images.py --file datasheet.pdf --output-dir ./figs/.
   • Identify which PNG files contain characteristic curves (I-V, gain-bandwidth, thermal derating, etc.).
   • Note the axis labels, units, and scale (linear / log) from each graph.

4. Digitize characteristic curves.

   • Run scripts/digitize_curve.py with axis calibration for each graph image.
   • Calibration: provide two known points on each axis in both pixel and data coordinates.
   • The script detects curve pixels by color signature, maps pixel coordinates to data coordinates, and outputs a CSV.
   • For multi-curve graphs: run once per curve with different --color argument.
   • Load references/curve-digitization.md for calibration methodology and accuracy limits.

5. Generate Python plots or export for simulation.

   • Run scripts/generate_curve_plot.py to recreate the graph in matplotlib from the digitized CSV.
   • Annotate with axis labels, units, title, and optional standard limits.
   • Export CSV for direct use in simulation scripts (e.g., as lookup table in SPICE or Python solver).
   • Load references/data-export-formats.md for SPICE table format and numpy-ready CSV conventions.

Reference Guide

Bundled Scripts

Dependencies: pip install pdfplumber pillow matplotlib

Constraints

MUST DO

• Preserve original units from the datasheet — report them in every output file header.
• Report estimated digitization accuracy (typically ±2–3% FS) when exporting curve data.
• Include source metadata (datasheet part number, page number, figure number) in output file headers.
• Validate that axis calibration points span an adequate range (at least 50% of image width/height) to minimize scale error.
• Warn when extracting from scanned datasheets — table extraction degrades significantly without a text layer.

MUST NOT DO

• Convert units silently — always ask or annotate when unit conversion is applied.
• Use image-space (pixel) coordinates in output CSV — always map to data-space values.
• Assume graph scale is linear without confirming from axis labels — log scales require logarithmic mapping.
• Report digitized data without axis calibration metadata — calibration must be documented for reproducibility.
• Skip the source image validation — always confirm the graph image is readable and axes are visible before digitization.

Output Template

For datasheet extraction tasks, report:

1. Component: part number, manufacturer, revision, datasheet page count.
2. Parameter tables: count, names (e.g., "Absolute Maximum Ratings"), pages, rows × columns.
3. Curves digitized: figure number/name, page, axis labels + units + scale (linear/log), number of data points extracted, estimated accuracy.
4. Output files: one CSV per table, one CSV per curve, plot PNG if generated.
5. Data quality notes: any tables with merged cells, any graphs with poor contrast or overlapping curves.

Primary References

• pdfplumber — https://github.com/jsvine/pdfplumber
• Pillow (PIL Fork) — https://pillow.readthedocs.io
• matplotlib — https://matplotlib.org
• WebPlotDigitizer — rationale and algorithm reference for manual curve digitization