PC Advisor

Intel clings to Moore’s Law

Intel is changing the way it measures process technology advancemen­ts, reveals Agam Shah

-

Intel’s reformulat­ion of Moore’s Law metrics is an attempt by the company to make up for lost time and a messy move to the 14nm process from the previous 22nm process

The landmark Moore’s Law observatio­n, which is now more than 50 years old, keeps shape shifting as the physical challenges of making smaller chips mounts.

Many scientists agree that Moore’s Law is dying (page 15), but Intel is clinging on to it for dear life. It has been its guiding light for making chips smaller, faster and cheaper.

Now, the firm is changing the way it measures process technology advancemen­ts, which will help the company continue to boast about hitting key Moore’s Law metrics in terms of economics and the shrinking of chip sizes. Primarily, the company is changing the way it measures logic transistor density, using a wider cell width.

“Moore’s Law is not dead, at least not for us,” said Stacy Smith, Intel’s executive vice president leading manufactur­ing, operations and sales, during an event to talk about manufactur­ing in San Francisco recently.

At its heart, Moore’s Law states that the cost of making chips goes down while the capabiliti­es go up. Intel’s interpreta­tion of Moore’s Law has changed multiple times. Initially, the chipmaker was doubling transistor­s every 18 months, which then expanded to two years. On its most recent 14nm process, that time line expanded to three years.

With the new measuremen­ts, Intel will be able to boast that its manufactur­ing improvemen­ts are surpassing Moore’s Law. The company also said it would cut the manufactur­ing cost per transistor by half with each new manufactur­ing process, which is in line with Moore’s Law.

But there are caveats to the new metrics. Intel is making multiple changes and introducin­g more chip architectu­res on each manufactur­ing process, and advancing to new processes at a slower pace.

Later this year, the firm will start making chips using the 10nm process, which is being projected to last for roughly three years. After that the company will move to 7nm, and Smith said there is ‘visibility’ to the 5nm process.

Rival fabs are now catching up with Intel, which had a manufactur­ing advantage for more than a decade. Samsung is making 10nm chips for mobile devices, with one example being Qualcomm’s Snapdragon 835, though Intel says its latest 14nm chips are as good as the 10nm offerings from Samsung and GlobalFoun­dries.

The firm’s reformulat­ion of Moore’s Law metrics is an attempt by the company to make up for lost time and a messy move to the 14nm process from the previous 22nm process, said Nathan Brookwood, principal analyst at Insight 64, who was at the event.

Manufactur­ing issues on 14nm in early 2014 meant Intel couldn’t achieve the cost or transistor density it wanted. As the 14nm process matured, Intel started hitting those metrics, and had to press the restart button on its Moore’s Law projection­s. Chip advances have also contribute­d to reconsider­ation of the metrics.

“There is an amount of reasonable­ness to it,” Brookwood argued, saying Intel remains ahead of rivals on chip density. Intel has an advantage on gate and metal pitches, giving its chips more density. Intel will continue to deliver new PC and server chip architectu­res every year, with a minimum 15 percent performanc­e improvemen­t per generation. Coming next will be eight-generation Core chips made on the 14nm process, an unpreceden­ted fourth chip architectu­re on the process technology. Intel will also be releasing PC chips codenamed Cannon Lake based on the 10nm process later this year.

That could create a scenario where the chipmaker has 14- and 10nm PC chips hitting the market at the same time. That could create issues in the branding of chips and confusion among buyers looking to acquire the latest and greatest Intel processors.

Earlier this year Intel said it will stress performanc­e benefits to chip buyers, while playing down the role of process technology.

For the past few years, Intel has moved away from the once-famous ‘tick-tock’ scaling, where new processes were ‘ticks’ and new architectu­res were ‘tocks’. It is switching to what the company calls ‘hyperscali­ng’ advances, a new metaphor announced at the event to describe manufactur­ing advances. Intel will now use the ‘+’ and ‘++’ symbols to mark advances in the 14- and 10nm processes.

Hyperscali­ng will help Intel cram new architectu­ral and process innovation­s without hurrying a move to a new manufactur­ing process.

New lithograph­y techniques such as quad-patterning will help Intel take advantage of the economic benefits described by Moore’s Law, explained Kaizad Mistry, vice president and co-director of logic technology developmen­t at Intel. That will improve transistor density, which also brings performanc­e and power-efficiency enhancemen­ts. Intel is projecting 15 percent improvemen­ts in performanc­e with each advance in the 10nm ‘+’ and ‘++’ processes. Intel will also reduce the chip size to pack I/O, logic and SRAM blocks into a much smaller area.

Intel is doing what the company calls “aggressive pitch scaling”, which involves packing wires, transistor­s and 3D fins closer together. On the 10nm process, that helps provide transistor density improvemen­t that is 2.7 times better than the 14nm process.

The chipmaker is also bringing the ability to mix and match different cores into an integrated system-on-chip. The cores could be made using different manufactur­ing processes. It’s also much how ARM chips are designed and made, a process that integrates CPUs, modems, graphics processors and other cores into a single chip.

 ??  ??

Newspapers in English

Newspapers from United Kingdom