Impact case study database

The University of Bristol spin-out XMOS Ltd. was created by Professor David May in 2005. Its headquarters are in Bristol, with offices in the US (New Hampshire) and Hong Kong.

The subject of the previous REF was the first generation of XMOS’ flagship product, the xCORE, the architecture for which is based on the University of Bristol research published in [3.1-3.5]. The IP that was spun out was a unique architecture designed to be intrinsically suited for real-time systems. The processors make it easier to design complex embedded systems for a range of consumer, audio, industrial and automotive applications. The same architecture from the previous REF has turned out to be ideal for new markets in voice processing, artificial intelligence (AI) and machine learning (ML), and so, in February 2020, XMOS announced a new product, the world's lowest-cost AI microcontroller [5.5]. It offers electronics manufacturers uniquely short development cycles, an agile build process, energy efficient processors, and the ability to correct faults quickly.

Economic growth of the XMOS spin-out / launch of xCORE-200

XMOS has seen significant growth during this REF period. In 2015, the company launched the second generation of xCORE, the xCORE-200. This was followed by the launch of xCORE-AUDIO and xCORE-VOICE in April 2016, and xCORE VocalFusion in June 2017.

xCORE-200 retained the first generation’s core architecture, and thus applicability for voice processing in edge devices (e.g. smart speakers, voice-enabled Internet of Things, etc.), while also adding new features to improve performance, such as the addition of a dual issue mode, which doubled the peak performance, quadrupling the memory on the chip, and doubling the number of cores, all of which enabled superior Digital Signal Processing (DSP) capabilities and provided the computational power for designing far-field microphones. Enabling the design of far-field microphones is a key advance, as the modality of interfacing with computers and appliances is changing from being button-driven (e.g. TV remote control, buttons on a cooker-hood) to voice-driven. Capturing voice in a cooker-hood requires picking out the voice from the noise of the fan. Capturing voice in a TV requires picking out the voice over the sound of the programme that is playing. The DSP associated with these two computations is significant and is something that the second generation XMOS device is capable of performing [5.1] by exploiting the innovative multi-threaded processors, instruction sets and deterministic performance developed in [3.1- 3.5].

xCORE-200 has enabled XMOS to grow significantly from the last REF period, from 50 to 70 staff, and in annual revenues from GBP4 million in 2014 to over GBP10M in 2020 [5.1]. XMOS has sold more than 10 million of these new devices to date, mostly in the prosumer audio market, with audio brands such as FocusRite, Sony, Meridian and Native Instruments forming a significant part of the number of devices sold [5.1].

Enabling voice interaction in new products across the global Voice and Edge-AI market

The high performance, low-cost and deterministic performance of the original XMOS architecture has proven ideal to tackle the emerging “voice recognition at the edge” market. Chips and software developed by XMOS are enabling voice interaction across a wide range of products.

AudioExpress (2017) stated that: “ The new XMOS VocalFusion 4-Mic Dev Kit features a compact four-microphone linear 100mm array. The captured voice signals are crystal clear even in noisy environments, enabling commands to be accurately captured from across the room for processing by the Alexa cloud-based speech recognition system. XMOS is the only solution to date that packs audio digital signal processing (DSP) and programmable I/O processing and control software into a single chip, enabling customers to significantly reduce their bill of materials, flexibly tailor their designs, and reduce costs.” [5.2]

The technology allows XMOS to offer their customers an integrated solution at a competitive market point. Meridian, a UK manufacturer of high-end audio products, has stated that: “ *Using XMOS we were able to replace three separate development systems, saving us over 6 months in development time. All our new products now use xCORE.*” [5.1]. Meridian’s products which have adopted XMOS include Meridian Explorer, Director and the Control 15 Media Server.

In 2017, XMOS’ VocalFusion technology was selected by Pegatron Corporation for its new smart voice assistant, Martina. The technology delivers clear voice capture, the ability to connect to Google Cloud services and language recognition in four languages. [5.3]

In 2018, Tymphany, a global premier audio ODM, announced the latest project in building a new Alexa Built-In soundbar incorporating XMOS’ VocalFusion far-field voice processor. The soundbar connects with the Amazon Alexa Voice Service (AVS) via WiFi and has an excellent far-field voice capture and barge-in performance from across the room – even when the soundbar is playing content at high volume and the commands are spoken softly. [5.4] Tymphany’s Senior Director of Technology Management states: “ We are proud of our team's efforts to realise this complex, groundbreaking and great performing soundbar reference design. We would not have been able to achieve this milestone without the awesome support of the Amazon Alexa Voice Service and XMOS development teams”. [5.4]

In 2019, Amazon certified the use of the XMOS VocalFusion kits for Alexa-enabled devices, one of very few products who have this qualification; see Figure 1. [5.6] The Director of the Alexa Voice Service at Amazon has said: “XMOS provides the thin-form factor and high performance needed […] to add Alexa in new ways to stereo home entertainment and wall-mounted AV equipment.” [5.7]. The Chief Evangelist for Alexa at Amazon, went on to say about XMOS “ *Having partners like you build hardware means we have a way for everybody to build a consumer device that you can talk to.*”

Figure 1 - XMOS Architecture qualified for use in Alexa [5.5]

Other XMOS customers include:

• Freebox, a large French TV operator who have the XVF3500 stereo-AEC voice processor built into their Delta product [5.5]

• Pillo Health, where XMOS voice technology underpins the 24/7 healthcare robot Pillo. The device delivers therapy reminders, health information, stores pills and reminds users of their scheduled medication doses. Emanuele Baglini, CTO of Pillo Health said: *“While designing our device, we were looking for the best solution to improve the voice capture accuracy to give our users a great experience. We tested the XMOS voice processor against other solutions on the market, and it outperformed them all for clarity, accuracy, distance, and direction.*” [5.5]

• Murata, where XMOS XVF3000 voice processor is built-in to their NAONA Array Mic unit, delivering the capability to capture speech and the direction of speech from a distance [5.5]

• Sharp, the Japan-based multinational who are using the XMOS VocalFusion XVF3500 voice-capture technology as part of their new Windows-based video conferencing and remote collaboration solution; see Figure 2 [5.5]

• Orange, where the XMOS VocalFusion XVF3000 enables their Djingo smart speaker [5.5]

• FiiO, the Chinese company producing high-quality music devices, includes XMOS in a range of their products [see, for example, 5.8]

Figure 2 - XMOS far-field voice-capture technology used in the Windows collaboration display, by Japan-based Sharp [5.7]

The third generation of XMOS’ products, still based on the same underpinning architecture, went into volume production in the second half of 2020. XMOS’ CTO has stated that “... the technology that was spun out of the University of Bristol 15 years ago has turned out to be so generic that it has propelled XMOS into Voice and Edge-AI markets that nobody had foreseen in 2005. We are looking forward to what the next decade will bring us.” [5.1]