Exploitation Plan

This document details the areas in academia and industry to which this project is naturally aligned and how the exploitation in these areas will benefit project members and the wider UK reputation and economy in such.

Contents

The Embedded Hardware and Software Market

The value chain for embedded software and hardware starts with human capital. In the UK, this is predominantly derived through university education. This project will generate positive academic and publicity results for UoB. They will in effect attract an increasing number of young software and hardware engineers to the discipline. As we will become world-leaders in energy-efficient compilation, UoB will gain a unique selling point.

In the commercial sector, the wide range of embedded products involve the combination of intricate hardware design with an increasingly large software stack to add functionality. Thus ALL hardware products require efficient compilation. The reality of this is that the majority of engineering effort now resides in software design, and an effective compilation chain can improve the time-to-market and efficiency of final products.

This project thus provides a feed-in at a fundamental level to all electronic systems designs and any products made as an aggregation of smaller systems (the mobile telephone is a good example of this).

Size of the Market

As well as the direct compiler market – businesses working on compilers, there are the industries further up in the value chain, who are additionally supported by this work. These are the silicon chip and silicon IP industries and then at the top is the global electronics industry.

There are no published figures for the size of the compiler market. However we can see the activity in one part of the market, the open source sector, because its activity is public.

During the time span 1 January 2012 through 31 October 2012, 459 individuals contributed to the GNU Compiler Collection (GCC) project. Of these, 197 made 10 or more contributions, and so may be regarded as active contributors. Most of those are professional compiler developers, either working in Universities or in industry.

Whilst GCC is the most widely used open source compiler, we are aware that the LLVM project has a similar level of activity (as can be seen from the attendance at their respective conferences). We also know that Microsoft Visual C is the world’s most popular compiler, and that there are many other niche proprietary compilers developed by companies like Intel and ACE Technologies. It is thus reasonable to suggest that GCC represents 20% of the professional compiler activity worldwide.

Compiler engineering is specialist and well paid. The total cost of employing a high grade engineer in industry is typically around $US500,000 per year, while in Universities, the figure is lower. We take a $US250,000 as an average.

This indicates the total amount spent directly on compiler development is around $US250 million or £150 million. As a sanity check we instead took $US150,000 as an average salary and used the ONS/BIS multiplier of 1.82 to determine the economic activity. This provides a figure of £170 million, and we want more of that business to be in the UK.Compilers are a critical part of the value chain supporting any processor chip or silicon IP company.

This is an area where the UK excels. Some of the world's largest chip companies have major R&D facilities in the UK, for example ST Microelectronics and Infineon in Bristol. In Silicon IP, the UK leads the world. The largest supplier is ARM in Cambridge (market cap £9.5 billion), while Imagination Technologies (market cap £1.2 billion) is third. Both ARM and Imagination are companies which illustrate the UK’s strength in this sector. They have grown strongly for the past decade as the following graph of their market capitalisation shows.

It is worth noting that, while ARM are now known for low-power processor designs, their original differentiation in the market, was as the first chip design to be supplied with a comprehensive software development suite (RealView) including compiler, debugger and IDE. Quantifying the size and trends of the embedded market is notoriously difficult as has been noted within the TSB itself.


Compilers are always developed in close cooperation with the chip designers. So strength in UK compiler technology will help to support a strong UK silicon chip and silicon IP industry.

And at the top of the value chain is the global electronics industry worth around $US1 trillion (Consumer Electronics Association, 2012). ARM's quarterly report in July 2012, announced they were supplying 2 billion designs per quarter. Other UK silicon design companies like CSR and Imagination have similar success in supplying the electronics market.That success relies not just on the excellence of the silicon designs, but of the infrastructure supporting them, of which the compiler is a key component. As a result, any improvements made in compiler optimisation through this project are exploitable and are of value to all members of this value chain.

The Case for Open Source Software Business and Research Methods

Open source software is everywhere. Your smartphone probably runs Android, when you browse the web you quite likely use Firefox to look at websites, most of which are running Apache.

Open source is also big business for both customers and suppliers. Google and Facebook's huge data centres are built with open source software. Red Hat is a US corporation worth nearly $5Bn, MySQL was sold for $1Bn, and IBM the world's largest patent holder makes more from open source software than it does from patents.

Open source software is free, both in the sense of not paying for it, and in the sense of freedom to use it how you wish. But as the corporate examples given above illustrate, there is still plenty of opportunity to make money.

Embecosm uses a service business model. Although the software we work on is open source (GCC, LLVM, Verilator, etc), it is immensely complex, and requires adaptation for different computer hardware. For most companies it is more cost effective to use our specialist skills in via sub-contract, even though, being open source they could develop the software in house.

This same business model is used by IBM's open source business unit. Since its founding in 2008, Embecosm has on average more than doubled its revenue each year, and expansion plans for the company will sustain this over the medium term.

Such approaches to research have already been proven in this field. In a precursor project with the same partners, we had much success with this approach and successfully engaged the GCC and LLVM compiler communities.

Benefit to Embecosm and the University of Bristol

As an open source project, both University of Bristol and Embecosm will benefit in terms of an increased reputation in this ever-important field. For Embecosm, the project will establish a reputation as the leading experts in energy efficient optimisation. The boost to reputation will allow Embecosm (and the UK generally) to be seen as the location where low energy compiler technology is to be found. This will generate new customers and business opportunities for Embecosm around the world. UoB is already widely recognised for its work on low power electronic system design. Its reputation will be greatly enhanced by its participation in this project, strengthening links to industry in this field and providing a feed through to teaching and new research opportunities.

Progress So Far

This project explicitly utilises open source processes to maximise the impact of this work. Our plan from the start has been to make the wider community aware of this project and its outputs early on. By creating engagement from new stakeholders, our aim has been to create a follow-on community who takes the techniques forward without the need for further financial support.

Events

A major part of raising the profile has been by appearing at and supporting events where MAGEEC can publicised. These are the major events where MAGEEC has been seen:

  • Wuthering Bytes 2014. Talk by James Pallister on measuring energy consumption. This was a community event, which was an opportunity to raise the profile of energy efficiency amongst smaller players in the UK embedded sector.
  • TSB Energy Efficiency Workship, January 2014. Hosted by Bristol University, with participation by the MAGEEC team and ARM, this was an opportunity for engineers to get hands on experience with energy measurement. With an informal "hackathon" style, this was a first for the TSB, and judged by those attending, from a wide range of companies and universities, to be a great success.
  • FOSDEM 2014, February 2014. On the first day, Jeremy Bennett and Kerstin Eder gave one of the main track talks Who ate my battery. On the second day, the MAGEEC team ran a stand showing the energy measurment hardware, and an all day workshop on energy efficiency, with participants from Embecosm, Bristol University, ARM and Cambridge University. The workshop included a hands-on session, at which, thanks to the support of the TSB, we were able to distribute 50 energy measurement kits.
  • EEC SIG Annual Event. Hosted at cfms in Bristol, this allowed all participants in the feasibility study program to present their work. Jeremy Bennett, assisted by Simon Cook presented MAGEEC (video, with a demonstration of the hardware measurement board).
  • Euro-LLVM, April 2014. Held in Edinburgh, Embecosm staff attended, and presented a poster on the work of MAGEEC.
  • Future World Symposium 2014. This is a showcase for UK electronics and computing technology run by the NMI, the trade body for electronics, computing and systems in the UK and Ireland. Embecosm's stand combined a presentation on MAGEEC with demonstration of the Adapteva Epiphany processor, capable of delivering 100GFlops in just 2 Watts, and for which Embecosm have developed the compiler tool chain.
  • GNU Tools Cauldron 2014. This is the flagship conference for the GNU compiler community, and this year was held in the UK at the Cambridge University Computer Laboratory, with support from the Knowledge Transfer Network. Jeremy Bennett presented MAGEEC (slides, video), and gave the first ever public demonstration of MAGEEC in action.

Future events include:

  • EACO Workshop 7, September 2013. A European-wide collaboration of energy researchers from academia and industry. Simon Hollis will present the MAGEEC project, with James Pallister demonstrating MAGEEC in action.
  • Innovate UK 2014. Embecosm will have a stand presenting MAGEEC and demonstrating it in action.
  • LLVM North America 2014. The MAGEEC team have submitted a paper, in which we propose demonstrating MAGEEC working for an ARM target.

Peer reviewed publications

To further increase awareness of such technology and in keeping with standard academic practices, a number of peer-reviewed academic papers demonstrating the feats achieved through this project have been published under the leadership of the University of Bristol. In keeping with the open approach of MAGEEC, thse papers have been published under open access license agreements, with pre-prints on arxiv.org.

The final output of MAGEEC will be a paper describing the entire project, authored by all the participants, and which it is intended will become the reference paper for the project.

Informal publications

MAGEEC has run a blog since starting. At the time of writing (5 September 2014) there have been a total of 15 posts covering all aspects of the project, an average of one per month.

Mailing list, IRC, wiki and social media

The MAGEEC mailing list has proved an effective way for the project members to communicate. It is now starting to acquire members from outside the project.

Similarly the wiki has been effective in collating all the information about the project. A number of external participants have contributed, for example in providing a Windows version of the energy measurement workshop.

MAGEEC has its own Twitter account (@MAGEECorg), and regularly tweets about the project. There are also MAGEEC channels on Google+.

Hardware

The energy measurement board developed for this project has proved an even bigger hit than expected. Part of this has been down to the strong support we have received from the Technology Strategy Board. However its performance (2 millions samples of voltage/current per second from each of 3 sampling points) hugely outperforms all but the most expensive equipment currently available, while its cost (approx £35 in small volumes) makes it far cheaper than any alternative, by at least two orders of magnitude.

We need to make a second manufacturing run, and the interest generated by this has encouraged us to invite other companies interested in evaluating this technology to join in that manufacturing run.

The design was completely open from the start, and this openness has encouraged others to take the board very seriously.

Employment and student recruitment

Since the start of this project, Embecosm has been able to recruit 2 more permanent staff of the highest calibre (both with first class honours degrees) as well as taking on an intern who was studying for his masters degree.

The project has proved very popular with Bristol University students, with two working on the project in summer 2013 and four in summer 2014, one of whom has gone on to start a PhD in the field at Bristol.

Customer engagement

There is of necessity some confidentiality required when talking about specific customer engagements. However Embecosm's business has continued to grow substantially during the period of the project. A number of customers have been provided with energy measurement boards for evaluation.

There is no doubt that Embecosm's participation in this project has led to increased customer engagement, renewal of existing contracts, and opportunities for new business.