Ceva rolls next-generation audio cores
4/16/2012 7:00 AM EDT
SAN JOSE, Calif. – Ceva has announced its next-generation audio DSP processor architecture and a family of four designs based on it. The Ceva TeakLite-4 consumes up to 25 percent less die area and 30 percent less power than the previous generation, the company claims.
TeakLite-4 can process a 256-point fast Fourier transform in 1,500 cycles and run at up to 1.5 GHz in a 28 nm high performance mobile process, the company said. It can be implemented in as few as 100,000 gates for a low-end version or up to 280,000 gates for its current top-end variant.
The core sports a customizable instruction set. It comes in versions with up to four 16x16, 16x32, 24x24 or 32x32 multiply-accumulate units (MACs). The cores also support customizable instruction sets, data caches and AXI interfaces.
“We don’t believe one size fits all anymore,” said Moshe Sheier, a product director at CEVA. “You need application specific cores and instructions for different market segments,” he said.
Low-end versions of the new cores will handle basic audio codec processing, typically in small standalone DSPs for handsets, tablets, PCs and TVs. High-end versions will tackle 32-channels of voice for multiplayer videogames, virtual surround sound processing for high-end audio subsystems and high-definition audio in a variety of systems.
High definition audio is already coming to 3G handsets with codecs supporting 16 KHz audio signals up from just 8 KHz used in traditional wired phones. Skype already supports a 24 KHz codec for high quality VoIP links on PCs. The 3GPP standards group is expected to define a 16 or 24 KHz codec for voice over LTE, said Sheier
Action has been brisk lately with new audio DSPs from Analog Devices, MIPS and Texas Instruments, said Will Strauss, principal of market watcher Forward Concepts (Tempe, Ariz.). The relatively low speed audio DSPs help save energy compared to processing audio, for example, on a faster baseband DSP, he said.
Currently, Nvidia, Qualcomm and TI all use their own audio DSPs in their smartphone chip sets. Broadcom, Intel, Mediatek, Morning Star, Samsung and Spreadtrum are among those who currently use Ceva audio cores in at least some of their chips.
Among Ceva’s four new products, the TL410 has the smallest die size and offers one 32x32-bit and two 16x16-bit MACs. The high end TL421 sports cache controllers and a master/slave AXI system interface to link to on-chip CPUs or DSPs.
Low power versions of the cores will be available for licensing before June. High performance versions will be available before September.