Samplify switches from data compression IP to the fabless semi business

Some time ago we wrote here about Samplify Systems, a company producing an unusual and quite efficient suite of data compression algorithms. Today, Samplify will announce a switch in business plans, from licensing algorithms to selling silicon. The company will announce an off-the-shelf product aimed specifically at the medical imaging market, with the hope of halo impact on the wireless base station market, test and measurement, and other areas that require transmission of high-speed data.

The product is a 16-channel, 12-bit, up to 65 Ms/s A/D converter with Samplify’s data compression integrated onto the back end. The result is a single chip that can absorb high-speed sensor data and produce a digital data stream while reducing the required number of LVDS pairs by up to 75 percent.

It’s not easy to say exactly how effective the compression is because there are several modes available, including a lossless mode, a fixed-ratio mode, and an on-the-fly adjustable mode that can be used to optimized compression subject so some other constraint, such as signal-to-noise ratio at the receiving end, in which case the effectiveness is somewhat pattern-sensitive. The 75 percent cited by the company is presumable in fixed-ratio mode.

In addition to the integrated data compression, Samplify is quite proud of the ADC itself, which was licensed from a third-party analog IP provider. The converter is said to use a full converter-per-port architecture, minimizing noise from crosstalk. Operating at 1.2 V in a 130 nm UMC process, the entire device consumes 44 mW per channel.

In addition to saving converter power, transmission power, and interconnect bulk, the device promises benefits at the system level. Since data from the ADCs normally flows into an ASIC or FPGA in imaging and wireless systems, compression that reduces the number of LVDS pairs necessary to carry the data also reduces the pin-count of the receiving chip, and potentially—if algorithms exploit the data compression—the memory footprint of data buffers. In addition, as Samplify has hinted in the past, since their compression scheme is essentially linear in nature, some DSP operations can be performed directly on the compressed data without need to re-expand it, substantially reducing the required bandwidth in the data paths.