Over the past half century, Moore’s Law has held up remarkably well—though nowadays it is more an industry target than a prophesy, and therefore a self-fulfilling one at that. Even so, the world has benefited handsomely from the decline in the cost of computing made possible by constant manufacturing improvements that have shrunk the width of semiconductor circuitry from tens of microns (millionths a metre) in the early 1970s to tens of nanometres (billionths of a metre) today.
But all good things come to an end. Engineers have long anticipated that Moore’s Law would cease to apply around 2015. By then, the components on a chip will have shrunk to a point where quantum-tunneling effects make it difficult for a processor to function efficiently. Quantum effects start cropping up when critical transistor dimensions become less than 15 nanometers (nm) or so. Intel, which effectively sets the standards for chip-making worldwide, will start replacing its existing 32nm process with 22nm technology towards the end of this year, with 14nm rules expected to follow around 2014. Without some fundamental rethink in chip design, future gains from shrinking circuitry would therefore be problematic.
Such a rethink has been underway for over a decade, and is finally ready to go into production. Earlier this month, Intel unveiled a radical three-dimensional chip architecture—the first big change in semiconductor layout in 40 years—that will be produced using its new 22nm process.
Since their inception, integrated circuits have had a two-dimensional planar structure, with a metal gate mounted across a flat conducting channel of silicon. Via its single contact patch, the gate controls the current flowing from the source electrode at one end of the silicon channel to the drain electrode at the other. However, the channel width has shrunk with every new generation of the technology—so more transistors can be packed into the limited space for greater performance. In the process, the gate itself has also become smaller and less effective, allowing current to leak away and impairing the transistor’s ability to switch rapidly between its two states.
To get around this performance roadblock, Intel’s new transistor design features a conducting channel in the form of a vertical silicon fence thatstands proud of the surface. That gives the metal gate straddling it three contact areas instead of just one to exert control over the current—a large patch on either side of the fence and a smaller one along the top. The result is less leakage and thus more current flowing when the transistor is in its “on” state (for higher performance). And thanks to the greater gate control, the current is virtually zero when the transistor is in its “off” state (for lower power consumption). Intel reckons chips using its “Tri-Gate” design can switch 37% faster than equivalent processors based on today’s technology, use 50% less juice and yet cost only 2-3% more to make.
The Tri-Gate design is to be used throughout Intel’s processor range—from powerful gaming and server chips to humble devices used in netbooks. By all accounts, the three-dimensional layout will work with 14nm as well as 22nm process technologies, ensuring Moore’s Law continues to apply for at least another four or five years. The design can also be tuned for high speed or low power consumption. For Intel, the latter is currently the most crucial requirement.
Intel has over 80% of the market for PC processors, but it is at the back of the pack in the race to power smartphones and tablet computers—the fastest growing sector of the computing business. Here, the front-runner is ARM, a British firm, which has a 95% stranglehold on some parts of the business. ARM, which does no manufacturing, licenses its processor designs to chipmakers around the world. What distinguishes its designs from other mainstream processor chips is their use of an advanced “reduced instruction-set computer” (RISC) approach pioneered by Acorn, another British firm, in the early 1980s. ARM (short for Advanced RISC Machines) was spun out of Acorn in 1990, to create low-power RISC processors for Apple and other customers. To date, over 15 billion ARM cores have been shipped by the company’s 200 or so hardware licensees (see “Send in the clones”, March 11th 2011).
What makes ARM processors ideal for the cramped innards of a handheld gizmo is their compact design, low operating temperature and frugalpower consumption. Owing nothing to Intel’s power-hungry x86 architecture, they incur no royalty fees to Intel. Nor do they need to be backwardly compatible with the x86 instruction set used by Intel processors and work alike chips from AMD, VIA and others. That is the key to the design’s low power consumption. The processor in Apple’s iPad 2, for instance, has a pair of ARM cores working in tandem to deliver ten hours of battery life between charges. Three out of five tablets now hitting the stores use similar ARM-based processors.
Intel would love to have a bigger slice of the handheld pie, especially now the Apple iPad and other tablets have begun to eat into sales of Intel-powered netbooks and even laptops. With annual sales of only $7 billion, the mobile computing market is still small beer by Intel’s standards—but the business is growing too rapidly to be ignored. Unfortunately, the most miserly Intel processors (the Atom range used in Windows-based netbooks) still consume between two and ten times more power than a typical ARM processor. However, a Tri-Gate chip made on 22nm plant would be a lot more competitive. No question that, if built on 14nm equipment, a Tri-Gate processor optimized for battery life would give today’s ARM chips a run for their money.
Challenging them is one thing, but displacing them in a different matter. Intel’s Tri-Gate chips would have to be not merely as good as ARM processors, but significantly better than them, to stand any chance of getting makers of portable devices to ditch their considerable investments in ARM technology and support. It also assumes that ARM and its 700 or so software and hardware partners cease innovating over the next couple of years. That is most unlikely. Indeed, apart from driving power consumption of its cores down below the one-watt level, ARM is now pushing into Intel’s performance territory with multi-core designs. IDC, a market research firm based in Framingham, Massachusetts, expects over 13% of the processors in PCs to be based on ARM designs by 2015.
Over the past week, a rumour has been doing that rounds implying Apple is about to abandon the use of Intel processors in its Macintosh computers. The scuttlebut suggests Apple wants to consolidate all its products on processors using the ARM technology used in its iPhones and iPads. Having little concern for backward-compatibility, Apple has made dramatic reversals in its hardware and software directions before—and will doubtless do so again. But it is hard to see Apple making such a move within the next few years, especially now Intel has revealed what its Tri-Gate processors can do, at least in desktops and laptops if not in handheld devices. There, at least, the ARM twisting looks set to continue.