Cadence Launches HiFi iQ DSP for Voice AI and Advanced Audio

Estimated reading time: 3 minutes

Cadence announced the Cadence® Tensilica® HiFi iQ DSP IP, the sixth generation of its highly successful HiFi DSP family, based on a new architecture purpose-built for next-generation voice AI and emerging immersive audio. As these applications proliferate in the home entertainment, automotive infotainment and smartphone markets, the HiFi iQ DSP delivers the performance and energy efficiency required to meet growing SoC computational demands. Compared to the industry-leading Tensilica HiFi 5s DSP, the HiFi iQ DSP offers 2X greater compute performance, 8X higher AI performance and more than 25% energy savings for most workloads—delivering over 40% performance uplift on numerous audio codecs.

As next-generation voice AI and emerging immersive audio applications proliferate in the home entertainment, automotive infotainment and smartphone markets, the Cadence Tensilica HiFi iQ DSP delivers the performance and energy efficiency required to meet growing SoC computational demands.

As next-generation voice AI and emerging immersive audio applications proliferate in the home entertainment, automotive infotainment and smartphone markets, the Cadence Tensilica HiFi iQ DSP delivers the performance and energy efficiency required to meet growing SoC computational demands.

“With the increased growth of AI across virtually every facet of the industry, audio and language have become critical interfaces for users,” said Anshel Sag, VP and principal analyst, Moor Insights and Strategy. “The AI performance and energy savings with Cadence’s new flagship Tensilica HiFi iQ DSP position it to enable future AI-enhanced applications and offload processing from other IPs.”

A number of factors are driving the need for greater computation, including the need for richer experiences enabled by the latest immersive audio codecs, audio processing at higher sampling rates, object-based rendering, seamless natural language processing (NLP), and automotive road-noise cancellation (RNC). These usages require increased AI/ML capabilities and enhanced audio/voice pre- and post-processing functions, which demand more computations performed with low latency and low power for on-device processing in edge AI and physical AI applications. Since most of these designs are power constrained, increasing the core’s operating frequency or integrating multiple cores is often not a viable option.

Leveraging more than two decades developing audio solutions and expertise in instruction-set optimization, Cadence has developed a new architecture specifically designed for AI-enhanced, high-performance audio workloads. The Tensilica HiFi iQ DSP addresses the latest audio market trends, delivering significant improvements over the Tensilica HiFi 5s DSP:

• 2X greater raw compute performance and 8X higher AI performance for compute-intensive advanced audio standards
• More than 25% energy savings across various workloads for very energy-efficient on-device AI processing
• Enhanced auto-vectorization for easy programming and fast time to market
• Integrated support for FP8, BF16 and other formats to run cutting-edge voice AI models

The Tensilica HiFi iQ DSP’s architectural enhancements and increased computational capability enable sophisticated immersive audio codecs such as Dolby MS12, Eclipsa Audio, Opus HD, Audio Vivid and others to run with higher efficiency than on previous HiFi DSPs. Cadence has already demonstrated more than 40% greater performance than the Tensilica HiFi 5s DSP on multiple codecs. Signal processing algorithms for keyword spotting (KWS), active noise cancellation (ANC), beamforming and automatic speech recognition (ASR) can be performed seamlessly, enabling state-of-the-art NLP functions. Additionally, multi-stream and multi-channel audio playback facilitates rendering of 3D spatial zones and sound bubbles, delivering highly realistic listening experiences.

“With recent advancements in LLMs, SLMs, energy-efficient SoCs and on-device AI, voice input is poised to become the new keyboard. To enable seamless interaction with the latest physical AI applications, SoC providers need energy-efficient and easily programmable IP capable of running SLMs on-device while performing voice and audio processing seamlessly,” said Boyd Phelps, senior vice president and general manager of the Silicon Solutions Group at Cadence. “As the market leader in DSPs for audio, voice, speech and AI applications with billions of Tensilica HiFi DSPs in a wide array of embedded products, SoC and system vendors rely on Cadence to continue to push the envelope for performance and energy efficiency.”

The Tensilica HiFi iQ DSP is fully capable of running popular small language models (SLMs) and large language models (LLMs) on the DSP itself, acting as an all-in-one AI processor for voice AI applications. For an additional performance boost, it can be easily paired with Cadence’s Neo™ NPUs or customer-built NPUs to further maximize performance and energy efficiency.

The Tensilica HiFi iQ DSP is compatible with the Cadence NeuroWeave™ Software Development Kit (SDK), TensorFlow Lite for Micro (TFLM), LiteRT and ExecuTorch environments for AI model execution. The Tensilica HiFi iQ DSP also leverages the vast ecosystem of partners and OEMs that have developed software libraries, compilers, codec packages, frameworks and more to enable easy deployment and fast time to market.