A Blog by Jonathan Low

 

Jan 4, 2015

Is Moore's Law Reaching Its Limit?

Eternal verities and all that. Moore's Law has defined the success and the hope of technology since Gordon Moore, the perceptive engineer who co-founded Intel theorized in 1965 that performance gains could be generated at ever lower costs.

This has been the bedrock promise of the technology industry and the root cause of the upheaval that has shaken economics and society ever since.

There are many who hope that nanotechnology and other advances in related fields will keep the dream alive. But 50 years is a long time in this civilization. And as the following article explains, there are challenges to the notion that this concept is forever. Some of them are technological, some chemical and physical. Some are also economic and political and sociological. It was the belief of the early post-war tech generation that they could do good by doing well. But as so often happens, success can breed a sense of entitlement which conflates personal with societal benefits.

Technological advances may sustain Moore's Law for some indeterminate period of time. But it's conceptualization was freighted with a number of assumptions. It's not that the Law itself is wrong or outdated or limited, but that the changes it has helped wrought in the society around it may contribute to its eventual end. JL

Brad McCredie reports in Venture Beat:

The reality is, the literal definition of Moore’s Law that every 18 months you can get the same number of transistors (or Silicon Functionality) for half the price is reaching its limit.
Is confidence in Moore’s Law dwindling? Is it dead already? Or did the concept actually exist prior to Gordon Moore’s coined theory and has continually evolved over time as it will continue to do long past the end of a silicon-centric era? No matter the stance, there is buzz in the industry right now about what the future holds.
The reality is, the literal definition of Moore’s Law that every 18 months you can get the same number of transistors (or Silicon Functionality) for half the price is reaching its limit. However, the concept of consistent cost/performance gains is very much alive through adaptation. Specifically, as silicon scaling alone no longer delivers cost-effective innovation and cannot meet the performance demands of cloud, big data, and analytics, the technology industry must rethink how to deliver on the promise of Moore’s Law. Continued progress will take a very different path from today’s results-challenged “tick tock” development cycle based on the technology scaling observed by Gordon Moore — and it will disrupt the hardware industry.
The “tick tock” approach of moving to a new chip manufacturing technology every two years and then updating the chip architecture but leaving manufacturing unchanged in the years in between is not meeting the silicon scaling requirements of today’s modern environment. Processor speeds aren’t increasing at the same pace anymore, and in turn it’s becoming quite expensive to get performance out of silicon. To expedite innovation, it’s time to look differently at the approach — move beyond processor-only design to take into account software, post-silicon materials and most importantly the benefits of an open, collaborative ecosystem.
A similar trend happened in the automotive and commercial aviation industries, where innovation agendas have expanded to support exponential advancements beyond those offered by their initial technology exploitation phases. For instance, in commercial aviation, from the Wright Brothers to the Concorde, the industry largely perfected flight times or aircraft speed. Now, as it has become very clear that technology cannot really support any further “speed” improvements, the industry looks to improve on the creature comforts that airline passengers have come to expect — from overall cabin design, like the 787 Dreamliner, to gourmet food options and personalized services. Similarly, in the automotive industry, when it comes to horsepower and engine design, generally, now that manufacturers have made it possible to travel over 100 mph, you aren’t going to see many differences in performance from a Dodge Challenger from the 1970s to today’s 2014 Dodge Challenger, for example. Advancement in automotive design now becomes more about safety and comfort amenities like airbags, satellite radio, and improved energy efficiency.
And for the IT industry, as the technology-driven components of microprocessor performance are hitting their technology asymptote of exponential improvements, a few forward-looking companies have come together to develop a new innovation model that disrupts the traditional “closed” nature of chip and server development in the past. The idea is to find innovations to carry forward the promise of Moore’s Law improvements through different technologies. Through open development of hardware and software and allowing intellectual property to be licensable to other manufacturers, the post-silicon scaling innovation era is already upon us.
The initiative is built around a new chip innovation that delivers the high performance customers need. By collaborating on new solutions that can be customized within data centers for new workloads such as cloud services or data analytics, the OpenPOWER Foundation is creating an opportunity to personalize systems to fit specific needs.
For instance, Nvidia developed a high-speed interconnect technology for linking its graphics processing unit (GPU) with IBM’s microprocessor, enabling data to move back and forth between them extremely fast. By opening up the licensing to the OpenPOWER Foundation members, it could lead to new systems that are up to 100 times faster than today’s.
Similarly, by opening up a new I/O standard called CAPI that enables communication with flash devices with a much lower software overhead, IBM and OpenPOWER partners Redis Labs, Altera, and Nallatech have created a NoSQL database server that can reduce storage costs by as much as 5x by replacing normal DRAM technology with flash technology at the same performance levels.
Moore’s Law is not dead, and I truly hope it never dies. I hope our industry will be able to innovate new ways to deliver that promise for a long time. Efforts are underway to discover what’s next post-silicon, such as carbon nanotubes, and non-traditional computational approaches, such as cognitive computing and quantum computing. In 10 years, it’s possible that completely different types of systems from what we know today could exist.
Right now we do not know what will be the winning technology to carry us forward. However, there are a few things we do know. First, no one company, alone, can spark the magnitude or diversity of the type of innovation we are going to need. Second, we are going to need innovation at every level in every discipline we can muster, to provide the required solutions in a post-silicon scaling era. Technology providers that collectively adapt and embrace bold, new ideas will not only survive, but thrive.

0 comments:

Post a Comment