Product Overview
Tensilica HiFi DSP Family
Configurable processors for audio, voice, and speech processing
Today’s audio, voice, and speech processing applications challenge designers to manage a wide breadth of performance and power requirements to create compelling, interactive, and immersive experiences for their customers. The Cadence Tensilica HiFi DSP Family for Audio, Voice, and Speech offers a low-energy, high-performance, highly optimized DSP solution that spans the entire spectrum of audio and voice algorithms and end equipment.
Overview
Overview
Audio/voice/speech (AVS) processing covers a wide range of performance- and power-consumption requirements. On one end of the spectrum is the ultra-low-power “wake-on-voice” processing in many of today’s smartphones and wearables. On the other end, building state-of-the-art voice-controlled digital assistants requires advanced audio digital signal processing capabilities to efficiently run neural network-based speech recognition. At the heart of these new AVS innovations is the desire of OEMs to create more compelling, interactive, and immersive experiences with their devices.
Using different processor architectures to handle the breadth of applications in the AVS domain would be very costly in terms of software development and product management. The Tensilica HiFi DSP family for audio, voice, and speech addresses this broad range of requirements, offering low-energy, high-performance processing for the entire spectrum of audio and voice-processing algorithms and end equipment while maintaining software compatibility across the portfolio.
HiFi DSPs are in SoCs across all major markets, including mobile, hearables, wearables, automotive, home audio, television, computing, and gaming (see Figure 1).
Whether you are optimizing to increase battery life, save thermal power, or save costs, there is a compelling HiFi DSP that meets your audio requirements.
With over 150 licensees, over 190 partners, and hundreds of software packages and growing, the HiFi DSP instruction set architecture (ISA) is the #1 DSP architecture for SoC designers.
Tensilica HiFi DSP Family
Auto-vectorization support
The HiFi 1s and HiFi 5s include special ISA and features for supporting auto-vectorization by the compiler. This greatly reduces the need for hand-optimization, thereby improving software engineer productivity and TTM. The same source code can be used optimally across HiFi 1s and HiFi 5s, without need for any per-DSP optimization.
Low-Power, High-Performance Audio and Voice DSPs
The Tensilica HiFi DSP ISA is highly optimized for AVS applications. By optimizing more than 300 software packages, the HiFi ISA provides the lowest energy and highest-performance DSPs for audio and voice. This performance scales across the entire DSP product family, from the ultra-low-power dual 32x16 HiFi 1 DSP to the energy-efficient 32x16 quad MAC HiFi 3z DSP to the high-performance 32x32 octa MAC HiFi 5s DSP.
Tensilica HiFi DSP Family Specifications
Features | HiFi 1 | HiFi 1s | HiFi 3z | HiFi 4 | HiFi 5s | |
---|---|---|---|---|---|---|
VLIW Slots | 2 | 2 | 3 | 4 | 5 | |
Fixed-Point MACs per Cycle | 32x32 | 1 | 2 | 2 | 4 | 8 |
32x16 | 2 | 4 | 4 | up to 8 | 16 | |
16x16 | 4 | 4 | up to 8 | up to 8 | 16 | |
NN MACs per cycle | 16x8 | - | 8 | - | - | 32 |
Audio and Voice Software and Ecosystem
OEMs continue to use AVS to differentiate their products and provide a compelling user experience, adding features such as voice triggers and advanced voice recognition. In the home audio and voice market, this could mean supporting the latest object-based audio decoder or the latest 3-D and AR/VR positional audio post-processing suites.
By choosing Cadence’s Tensilica HiFi DSP family, SoC designers can ensure that the broadest set of codecs and pre-/post- processing software is immediately available to their customers. More than 300 different software packages are currently available as part of the largest audio and voice ecosystem of any DSP in the market. All the newest and most innovative audio and voice IPs are always available with HiFi DSPs.
HiFi 1s and HiFi 5s also support imaging ISA for lightweight vision applications. These DSPs also have special MACs for ML inferencing for edge-AI and on-device AI applications. They also bring double-precision acceleration for audio algorithms that cannot tradeoff precision for range.
Flexibility
The Tensilica HiFi DSPs are highly configurable, giving the SoC designer control of numerous pre-defined functions and features, including the memory subsystem, debug, floating point, and many others.
In addition to configuration options, the SoC designer can choose to improve performance further and reduce energy for any given application by adding custom instructions and more I/O bandwidth. This is possible because all HiFi DSPs are based upon the Cadence Tensilica Processor Generator, which allows designers to add value through additional customization while retaining complete compatibility with the HiFi software ecosystem (see Figure 2).
Ease of Programming
A key advantage of the Tensilica HiFi DSPs is their simple programming model. Software developers can write audio and voice applications completely in C using efficient and optimized audio and voice APIs, matching or surpassing the performance of the same applications built with hand-code assembly. HiFi 1s and HiFi 5s support auto-vectorizing by the compiler, eliminating the need for time-consuming hand-optimization by DSP experts and allowing all embedded programmers to target the DSPs. This allows code to port seamlessly and optimally between HiFi 1s and HiFi 5s.
Additionally, Cadence offers comprehensive, highly optimized DSP and AI libraries that dramatically speed up the development of very efficient algorithms.
At the system level, Cadence offers XAF, a framework that allows customers to easily create and run a variety of audio and AI processing chains on HiFi DSPs. The framework provides a simple API for integrating audio components and modules and includes support for RTOS (Cadence XOS or FreeRTOS).
Customers can also leverage Audio Weaver from DSP Concepts. Audio Weaver is a drag-and-drop graphical UI with real-time tuning and debugging capabilities that enables users to create, tune, and productize their audio and voice processing features quickly and effortlessly.