Gerard Andrews, senior product manager with Cadence's audio/voice IP group, said: "Semico has predicted there will be 40billion sensors a year by 2018, which means more data needs to be processed, cleaned, filtered and acted on. In most cases, the devices which will be doing this will be low cost, with low energy requirements."
Cadence sees three factors as being important in IoT devices: sensing; computation; and communications. "The DSP workload associated with these three will increase," Andrews noted, "which is why we are taking a DSP centric focus."
The Tensilica Fusion DSP combines an enhanced 32bit Xtensa control processor with DSP features and flexible algorithm specific acceleration.
Using the Xtensa Processor Generator, IoT device designers can configure the options they need to create an optimised Fusion processor, which can be designed into SoCs for a range of applications.
The configurable elements include floating point support, up to four 16 x 16 MACs and AES-128 encryption.
"In the past," Andrews continued, "vertical DSP products have focused on specific domains. With Fusion, we want to excel in audio, voice and communications, so it's more of a general purpose DSP, with modules running voice and comms stacks." He added that voice control and recognition was featured as Fusion is likely to be used in wearables, where keyword detection is important.
Typical applications for Fusion are likely to be in products with low to medium compute requirements.
