Arithmatica, a startup concentrated on using advances in silicon math1 algorithms to increase speed and/or lower costs for math1-intensive integrated circuits, revealed that NVIDIA, Xilinx and Layer N companies utilized the firm’s advanced technologies to design their next-generation products.
Silicon intellectual property of Arithmatica helps companies to significantly improve silicon efficiency far beyond what can be done with traditional EDA tools, without changing current design practices. Applying it to math1-critical blocks in graphics chips can reduce overall chip area by up to 10%, saving millions of dollars in manufacturing costs for each design. It also improves performance in processor designs in varying degrees, depending on the application.
Compared to traditional methods, CellMath functions are based on faster carry-propagate logic that is central to efficient addition, faster parallel counters, core to multiplication, and optimizations in floating-point and single-instruction multiple data (SIMD) computational operations. CellMath also improves productivity and lowers risk by providing simulation models that are formally verified against the gate-level netlist. Companies can take advantage of CellMath without changing their design flow, standard cell libraries or process technology.
The CellMath technology does not depend much on the technology process, though, Arithmatica’s specialists are working closely with the company’s clients to tailor the tech for particular products. SiliconStrategies reports that there are works on a 65nm chip from an unknown developer.
Companies including NVIDIA Corporation, Xilinx and Layer N Networks successfully have incorporated CellMath technology in their flagship products, achieving impressive performance and cost goals, according to Redwood City, California-based company.
“The PC graphics community demands a constant pace of improved quality of experience at very competitive price points. Using Arithmatica’s CellMath Graphics Library, we were able to achieve a new threshold of performance in our next-generation of graphics processors while reducing the chip area dedicated to calculations in many of our major blocks by typically 20-30%. As an early customer, we worked closely with Arithmatica’s engineering team and we were able to embed their silicon IP in a variety of designs without any schedule impact within our projects. The result is that our new product line has the highest performance GPU in the market today and we are positioned to be very aggressive in terms of expanding our market leadership. It is refreshing to work with a young company that delivers on its commitments and we look forward to a long working relationship with Arithmatica,” said Gopal Solanki, Vice President Platform Products for NVIDIA.
It is not clear which processors of NVIDIA incorporate the new technology. Last year the company suffered from low gross-margins as a result of an un-mature 0.13 micron fabrication process of TSMC, high transistor count on the company’s GeForce FX product line and some design issues with its chips. This year the Santa Clara, California-based graphics company’s IC designs became even more complex and it seems to be fully natural for NVIDIA to employ a technology that would simplify the designs. It may emerge that the whole GeForce 6-series (NV4x-family) of graphics processors already uses the CellMath capabilities to boost efficiency of integrated circuits, however, NVIDIA has never announced this.
The CellMath graphics library, processor library and configurable instances are available now for licensing, priced from U.S. $175 000; Arithmatica’s licensing model includes a project-based, non-royalty fee pricing structure, eliminating periodic time-consuming accounting and payment procedures.