Tak’Asic JPEG IP is a part of Tak’Asic know-how in image compression/decompression. This Virtual Device is fully compliant with the baseline of the ITU T.81 standard, better known as JPEG. Its coding algorithm is a lossy image compression method. It means that it is mainly dedicated to compression of natural images in which loss of accuracy cannot be perceived by human vision.

Tak’Asic JPEG IP is an open macro-function and is able to integrate many options as byte rate control, sampling, filtering, color conversion and so on. Tak’Asic JPEG IP is available in three formats: a fully synthesizable VHDL source code, a generic RTL netlist or as a core function in a targeted technology. Ask Tak’Asic for availability of a silicon proof.


  • Full compliance with the JPEG baseline ITU T.81.
  • 8 bits/channel pixel depth.
  • DCT/IDCT with 12 bits fixed coefficients and 16 bits precision computation.
  • 2 programmable quantization / dequantization tables.
  • Programmable quality coefficient for bit rate control.
  • ZigZag sequence and Run-Length coding/decoding.
  • Bit rate measurement.
  • Huffman coding and decoding.
  • Programmable Huffman tables (two for AC coefficients and two for DC coefficients).
  • Support for restart markers.
  • SOI and EOI markers.
  • All other segments/markers parsed by software.
  • Support for 3 channels/component color.
  • Support for single component (greyscale).


  • Throughput : minimum 66MB/s in 0.35um technology (66Mhz ).
  • Up to 100MB/s in 0.25um technology.


  • Generic system interface for transfer synchronization.
  • 4-channels : Compression in, Compression out, Decompression in, Decompression out.
  • Byte/clock transfers (burst mode).
  • Parallel loadable table.



  • Automatic byte rate control using quality coefficients and byte rate measurement.
  • Byte rate control using variable quantization table.
  • Automatic byte rate control using FIFOs (Compression out).


  • 4:2:2, 4:1:1 horizontal sub-sampling only.
  • 4:1:1 horizontal and 4:2:2 vertical sub-sampling.


  • Programmable 3×3 matrix multiplication (linear conversions).
  • RGB to CIELAB non-linear conversion (to be defined).

These IP can be provided with JTAG and partial/full scan for testing purpose.

Download specification (pdf file 207 Kbytes)