Next Generation Networks (NGN) asked Professor Rudy Lauwereins, vice president of Belgium’s Interuniversity Microelectronic Centre (IMEC) for his thoughts on mobile handsets through 2012.
Professor Lauwereins heads one of IMEC's four research divisions. His group develops enabling technologies for consumer and battery-operated devices in the nomadic and mobile arena.
This is the first of a two-part interview.
NGN: Looking at current handsets, they do everything: they are cameras, MP3 and video game players, as well as providing web browsing and email. What will handsets deliver by 2012, and what will be new and novel?
RL: This is a difficult one. Today's phones deliver all you can imagine. But that doesn't mean there won't be new applications. What I can say is that screens will have higher resolution and probably be bigger in size, plus there will be much more storage. Devices will thus support more realistic games, higher (data) throughput and much more compute power. All these will make several applications possible but I have no clear idea.
NGN: What specification will they have and how will they compare to today's phones in terms of processing power, storage, and technologies within the handset? One example you have mentioned in the past is that handsets will have more than one camera to create views from any angle.
RL: I have been tracking the cost of compact flash (non-volatile memory) for the last eight years: what capacity memory you can buy for Euro 50 and what for Euro 1000. It is doubling every year. At the end of 2006, you can buy 16 Gbyte for Euro 1000 and 2 Gbyte for Euro 50. Memory companies tell me they expect this to continue for the next five years. That means handsets - at least high-end ones - will have 64 Gbyte in 2012. For low-end phones it will be 16 GB. What applications will use such cheap storage? Current cameras provide stills and poor quality video clips. By 2012 you'll have a video camera with you all the time.
Handset devices in 2012 will also have multiple processing cores. Handsets now have multiple cores but having had discussions with nine companies recently, the expectation is that by 2012 there will be between 16 and 100 cores per handset. That's ten times greater than today. There will also be silicon-scaling benefits [using smaller feature CMOS processes which will further boost each core's processing performance] so the overall processing performance improvement will be greater than ten times. It is hard to say exactly but maybe a factor of 30.
As for the multiple cameras, all nine companies say they don't know whether there will be a suitable business model for it and hence it is not a strong focus. Where there is pressure is to combine video and 3D graphics on one platform. High-end phones have a graphics-processing-unit now. There is a need to find an architecture that deals with video and 3D graphics without needing a dedicated [hardware] unit. The video and graphics will be used mainly for games but also for graphical data such as navigation maps.
NGN: What are the leading technical challenges to be overcome to make such handsets possible? For a start, Nokia and Qualcomm both mention that handsets will need to support eight distinct radio standards. Then there is the issue of growing power consumption while the energy capacities of batteries are not advancing in lock step.
RL: Sofware-defined radio will be in products in 2009 and the work on that is almost done, so this is almost in the past! What is being finalised now is passive software-defined radio – a flexible platform that can implement a number of radio standards but they will be ones I select when I want to use them.
The next step, to appear by 2012, is agile radios. This is a much more active form of software-defined radio. Here the handset scans to select the network available based on the user profile, such as what is lowest cost or what offers highest voice quality. But the terminal rather than the operator does the selection.
Then there is cognitive radio where the terminal scans the whole spectrum, finds a frequency it can use and then chooses the most appropriate modulation scheme. But a true cognitive radio is beyond 2012.
Batteries aren’t scaling well but then current phones, with batteries weighing 30g or 40g, last twice as long as previous ones, so we can solve the problem at the architecture and application level.
With 3G handsets it’s a new problem again but there is quite some innovation here in processing technology, algorithms and storage. All are used to increase the compute power for the same energy capacity.
The idea that by 2012 there will be new fuel cells that offer a quantum leap in energy stored is not seen as likely, but current batteries are increasing in energy capacity by 5% to 10% each year. We can certainly live with it.
Companies also points out that even if larger batteries emerge, the instantaneous power consumption will not rise above 3W. You put a phone in your pocket and it cannot go above the 3W mark for temperature reasons.
There is also the issue of programming the software code onto a multi-core architecture. The C [high-level language used to write the software code] compiler needs to be aware of the multi-processor architecture.
Then there is the issue of predictability. One application mapped onto the hardware may run in real-time and so may a second but when you run both? The mix of applications changes over time and guaranteeing that all the combinations will run in real-time is a problem. Ensuring all the applications in all combinations working correctly is infeasible. There are stories that after mapping applications onto the hardware correctly, it took 100 man-years of effort to solve the predictability stuff.
Another issue is scalability. Cell phone makers support some 60 different phone models using several general hardware platforms ranging from low end to high end. Since mapping applications is a huge job, they don't want to spend all this effort on each of the platforms. Rather they want to develop them once and get them running quickly on these platforms.
One last challenge, that IMEC is expert in, is building a reliable silicon platform with predictable performance, when going to smaller CMOS processes results in increasing transistor variability and decreasing reliability.
The second part of the interview covers fixed-mobile convergence, how to ensure designs of 2012 meet the fast-changing requirements of wireless, and what will 4G data rates be used for.
No comments:
Post a Comment