Donofrio said that the design flexibility enabled by processors and memory, together with modern-day transistor budgets and the design costs associated with making use of these transistors, make "embedded systems based on processors and memories the approach of choice going forward in this industry".

While the miniaturization of CMOS devices cannot, by itself, continue to yield the expected power-performance growth, the introduction of novel materials and device structures combined with improvements in power-efficient design will enable additional CMOS generations, said Donofrio. He predicted that the trend toward wireless, interconnected, embedded devices will not only drive growth in the embedded marketplace but will place increased demand on a range of servers that create and distribute value information on the network.

Citing the evolutionary path of brainpower over ages, Donofrio predicted that in less than 20 years the growth of compute power will theoretically correspond to hundreds of millions of years of human intelligence evolution. But it will take some doing getting there.

While much of the reduction in power-per-operation in modern chips has come from CMOS scaling, frequencies and chip capacitance have increased with every generation. And that Donofrio will hit a wall unless "we can simplify the processor core architecture, and reverse the trend toward ever-more complex and less power-efficient cores."

"To maintain performance growth with stunted per-core and per-thread performance growth rates, we must have a more rapid increase in the number of cores per die," said Donofrio. This places unprecedented requirements on the corresponding growth of "off-chip" bandwidth. Donofrio said that off-chip signaling frequencies are likely to exceed the frequency of processor cores in the not-too-distant future. "By increasing the number of cores, I'm convinced that a one-Teraflop multiprocessor die with one-Terabyte of "off-chip" bandwidth is feasible — at reasonable cost — before the end of this decade.

New materials, such as high-K dielectrics for the gate, and low-K dielectrics between interconnect, modifications of the substrate materials such as straining the substrate, or producing substrates with dual crystal orientations to enhance both electron and hole mobility, and new device topologies — are all technologies that need to be applied simultaneously to achieve that 1-Teraflop multiprocessor die.

Donofrio was optimistic that the needed technologies will be developed in time: "After 37 years in this industry, I have learned that absolutely nothing is the ultimate."

He cited developments from IBM Labs to support his case. Scientists at the Almaden Research Lab in San Jose have constructed the world's smallest working computer circuits (12 by 17 nm) by deploying carbon monoxide molecules on an atomic copper surface arranged, more or less, as a sequence of toppling dominoes. "The technique, called molecular cascade, explores the far reaches of science to find ways to harness the quirky behavior of atoms, molecules and quantum spins as alternatives to silicon."

Using this technique, scientists were able to build a typical six-transistor SRAM cell in 90 nanometer technology occupying a square micron and working digital-logic elements about 260-thousand-times smaller than those used in today's most advanced semiconductor chips.

In the end, pervasive access to unlimited and deep amounts of data and having computational capability or applications without actually having to worry about where they reside is where the industry is heading, according to Donofrio.

On the "pervasive" end, there are now more devices than people communicating through the Internet. "It's not a great leap to soon envision trillions of devices all connected, either directly or through wireless technology, and transmitting and extracting ever-more information," said Donofrio. "It's the aim of this industry to enable this computing paradigm in an uncomplicated way — in a way that's invisible to the end user, but very tangible to businesses and institutions everywhere."

And on the "deep" side — as computational power continues to increase — so does the ability to extract new knowledge and information. "More and more, as we go forward, the technologies we invent will be the technologies that hide [themselves], driving us to an incredibly connected, natural IT environment."