Monday, May 9, 2011

Intel Enters the Third Dimension with Radical Chip Design


This week, computer chip manufacturer Intel announced that it is preparing to enter a new dimension in transistors--literally. Known as Tri-Gate, its new transistor will be the first to go into mass production with a truly three-dimensional (3D) structure.
Intel says that the transistor will offer performance and efficiency benefits over 2D models when becomes production ready in a new range of microprocessors later this year. But with some industry analysts calling it a risky venture, Nature explores what the advantages are of 3D.
What do transistors do?
Transistors are the building blocks of microprocessors, which are the 'brains' or computational devices inside PCs, laptops, smartphones and pretty much all modern electronic devices. A transistor is essentially an automated switch that can store information as either a "1" or a "0", depending on whether the switch is on--letting electric current through--or off. The wiring of several transistors together creates a device called a logic gate, which takes these ones and zeros and performs basic calculations with them. Home computers available today contain billions of transistors wired into logic gates, and have huge processing power as a result.
What's the difference between a 2D and a 3D transistor?
Transistors are usually made from silicon, which is a semiconductor--a material that can behave as both an electrical conductor and an insulator. They consist of a straight channel connecting a source to a drain, interrupted half way by a wide gate. The gate is what makes the transistor a switch: apply the right voltage and a conductive pathway known as an inversion layer forms, allowing current to flow from the source to the drain. In this instance, the transistor is on; without the inversion layer, no current flows and the transistor is off.
All transistors mass-produced during the past 50 years or so have been 2D. This means that the source, the drain and the channel connecting them all lie flat on the same plane. In Intel's 3D transistors, on the other hand, the channel protrudes from the surface in a ridge or "fin". The result is that it has not one, but three sides in contact with the overlapping gate--thus its name "Tri-Gate".
Why is the 3D design better?
Chip manufactures such as Intel have been progressively shrinking transistors in order to pack more onto each chip and, ultimately, make faster computers. At present, the fastest chips use transistors that are about 32 nanometers in diameter--that's on the order of about 100 silicon atoms--and manufacturers will soon be producing 22-nanometer versions. But this continued miniaturization has an attendant problem: as the transistor's source and drain get closer together, and as the channel gets smaller, it becomes harder for the gate to control the formation of the inversion layer. Simply put, the distinction between "off" and "on" becomes fuzzier.
Having a 3D structure solves this problem. Because it is in contact with three sides of the channel, the gate has much greater control over the inversion layer. This means that the on and off states are more distinct even when the transistor is shrunk.
What is the benefit for computing?
Intel will be incorporating Tri-Gate structures into its next generation of 22-nanometer transistors, slated to be production ready by the end of this year. The company says that, compared with its current 32-nanometer, 2D transistors, Tri-Gate transistors will be up to 37 percent faster. In general, the 3D design should allow transistors to be packed more closely to one another, and so make it possible to fit more into the same space.
There are subtle advantages, too. According to Intel, the structures leak less current than standard 2D ones when not in use, which will improve the battery life of portable electronic devices such as laptops and smartphones. Moreover, when they are run at relatively low voltages, they should consume less than 50 percent of the power required by Intel's current transistors--which will be a boon for heavy-duty network servers.
The concept of 3D transistors has been around for well over a decade--the difficulty has been how to engineer one that can successfully be mass-produced. Because transistors now comprise just a few dozen atoms, even minor defects can have a huge effect on performance. Having features protrude from the substrate is particularly tricky, and Intel has not disclosed how its engineers have tackled this.
The Tri-Gate design is essentially a variant of a "FinFET" 3D structure developed in the late 1990s by Chenming Hu and his colleagues at the University of California, Berkeley. Other chip manufacturers such as IBM, Samsung and TSMC are all working on 3D designs, but are not expected to put them into production until at least the next generation of miniaturization, after 22 nanometers.
What will follow 3D?
Intel thinks the Tri-Gate structure should scale down to transistors of 14 nanometers and smaller. In principle, it should be possible to make transistors, 3D or otherwise, of just a few atoms, although manufacturing consistency becomes ever more difficult as size diminishes.
At some point, manufacturers will be forced to explore yet more dimensions. Perhaps at that stage the answer will be spintronics--an emerging technology that makes use of an electron's spin, as well as its charge.

No comments:

Post a Comment