Santa Cruz, Calif. - Promising to speed Matlab floating-point simulation by as much as 200 times, Catalytic Inc. Monday (July 18) rolled out the Catalytic Rapid Matlab Simulator. This third-party tool runs inside Matlab, an analysis and visualization tool provided by The Mathworks Inc.

Catalytic RMS incorporates technology from DeltaFX, Catalytic's first product, which facilitates the conversion of floating-point models to fixed-point in Matlab.

While DeltaFX targets software developers, Catalytic RMS is primarily aimed at hardware designers, said Lisa Schmidt, vice president of marketing at Catalytic. "We found that there's a vast amount of interest in Matlab for hardware design and that people really needed simulation acceleration in the Matlab environment," she said.

In the recent EE Times/Deutsche Bank 2005 EDA survey, 32 percent of the chip designers and 38 percent of the pc-board designers who responded said they use Matlab, and The Mathworks topped the user satisfaction survey among board designers (see June 13, page 24).

To reach out to hardware designers, Schmidt said, Catalytic RMS extends that company's coverage of the Matlab language to include such constructs as structures and globals. It provides even faster simulation than DeltaFX, she said, along with support for Linux and distributed workstation simulation.

"Another major difference is that people doing software design are building things in 8-bit increments, but hardware designers need variable bit widths, so Catalytic RMS supports that," Schmidt said.

Matlab, Schmidt said, is a very rich language, but it's an interpreted language that doesn't offer fast simulation. Catalytic takes a compiled approach that promises to speed floating-point Matlab by a typical 10 to 50 times, with acceleration above 200 times reached in some examples. For fixed-point simulation, the company claims a 100x to 1,000x times speedup.

"We're the same thing for Matlab that [Synopsys] VCS was for Verilog," said Catalytic CEO Randy Allen. "Matlab is an interpreted language that we accelerate by compilation." There is no loss in accuracy for doing so, he claimed.

Catalytic's technology compiles the Matlab language into a dynamically linked library. That, in turn, is converted into the C language. Users can link the compiled code back into the Matlab environment and execute it. "It's transparent to the user," Allen said. "It just runs 10 times faster."

Users still have access to the graphing and plotting capabilities offered by Matlab, Allen noted. Since Matlab programs typically don't involve large amounts of code, compilation time is "trivial," he said.

While the simulation run is a batch process, users have some control over the warnings that will be issued and the debugging that will be available. Users can check for overflow or underflow, track subscript ranges and set up code coverage.

Catalytic RMS can automatically determine which subfunctions can be easily revved. Through the grahpical user interface, users can provide information on subfunctions that will assist with the acceleration.

Catalytic RMS speeds designs that use the Matlab Signal Processing and Communications toolboxes. It supports most Matlab language constructs. "Some things are just very dynamic and interactive in nature, and you can't support them in compilation environments," said Allen. "But [in regard to language constructs], we haven't found anything that people significantly use."

Finally, Catalytic RMS provides fixed-point analysis tools that display the variables in a Matlab design, provide quantization recommendations based on simulation runs, and flag overflows and other potential quantization errors. The product is available now in Windows or Linux platforms starting at $15,000 for an annual license.