计算机专业时文选读(980)

Subatomic properties will remake computing

Imagine a data storage device the size of an atom, working at the speed of light. Imagine a microprocessor whose circuits could be changed on the fly. One minute, it would be optimized for database access, the next for transaction processing and the next for scientific number-crunching.

Finally, imagine a computer memory thousands of times denser and faster than today’s memories. And nonvolatile, so it retains its contents when the power is off.

All of these and more are on computing’s horizon, thanks to the exploding field of spintronics. Spintronics isn’t entirely new. The spintronic effect called giant magneto-resistance was introduced by IBM in 1997 in its GMR disk-read head. As a result, disk capacities have jumped by a factor of 100 in the past five years.

Electronic circuits are driven by electron flows, which have a charge that can be measured and controlled. But electrons not only flow; they also spin like tiny bar magnets. Depending on their orientation, the spins are said to be “up” or “down.” This additional variable, or “degree of freedom,” means that electrons can do more things and convey more information than they do in conventional electronics.

The most immediate research goal is to produce magnetic random-access memory (MRAM), which stores data using magnetism rather than electrical charges. Unlike the dynamic RAM in your PC, MRAM is nonvolatile.

IBM is working with Munich-based Infineon Technologies AG and says it will have MRAM in production as early as 2005. It will be 50 times faster than DRAM and 10 times denser than static RAM, and it could eventually replace both.

Others have even suggested that MRAM might replace disks for data storage. Putting logic and storage in a single chip would eliminate the slow disk I/O that’s a bottleneck in most computer processing.

IBM’s MRAM will use magnetic tunnel junctions, an application of spintronics in which electrons are allowed to “tunnel” between two ferromagnetic layers based on their spin. Each junction can store one bit. It promises a sort of universal RAM with very high performance —— high writing and reading speeds —— plus very high density and nonvolatility.

Further out, researchers are working on still more exotic applications of spin. David Awschalom, director of the Center for Spintronics and Quantum Computation at the University of California, Santa Barbara, is looking at what might be done with the spin of an atom’s nucleus, a new idea.

“The subatomic part of the atom would store the information, and the electron would act as the bus to carry information in and out of the nuclear subsystem,” Awschalom says.

He aims to build an optical-based information processor in which beams of light would transfer information to the nucleus through electrons. Such nuclear memories would be “many orders of magnitude” denser and faster than traditional semiconductor memories, he says.

Indeed, more broadly, the thrust of spintronics research will be to combine electronics and photonics with magnetism —— which traditionally involves metals —— in semiconductor materials. That will enable ultrafast and ultraefficient submicron devices that integrate computing, communications and storage. The slow interfaces between different materials that convert one kind of signal or property into another would be gone, and the latencies that typically slow the movement of data from one processing stage to another would be greatly reduced.

“You’d have everything integrated in a much simpler circuit,” says DARPA’s Wolf. “They would be much like existing semiconductor devices, except the current is spin-polarized.” That would enable, for example, the construction of very fast communication switches. “You could call it spin photonics,” he says. “They can easily operate at terahertz speeds.”

A semiconductor device can’t use spin until a way is found to get spin-polarized electrons into it, and that has proved difficult. But IBM recently demonstrated that it can use magnetic tunnel junctions to inject the current, as they do for MRAM.

IBM’s Parkin says spintronic semiconductors could be used to build reconfigurable logic devices. “So maybe your computer could be optimized for certain instructions by rearranging the way [logic] gates are connected, on the fly,” he says.

Another tough challenge has been to create magnetic semiconductors that sustain their spin states at room temperature, but physicists, materials scientists and engineers have made tremendous progress on that front just this year. The rapid development of spintronics seems likely to continue. The theory is in quite sound shape. There are many challenges, though.

亚原子特性将重构计算

想像一下数据存储装置只有原子大小、并以光速工作。再想像一下微处理器的电路能飞快地修改。只需一分钟，就完成对数据库访问的优化，接下来对交易处理优化，接着再对科学计算优化。

最后再想像一下计算机的存储器比今天的存储器的密度和速度都要提高几千倍。而且是非易失的，因此断电时它仍能保存内容。

由于旋转电子学研究的爆炸性进展，所有这些以及更多的新技术已经出现在计算的地平线了。旋转电子学不全是新东西。早在1997年IBM在其GMR磁盘的读出头中引入了称作巨型磁阻的旋转电子效应。结果在过去的五年中，磁盘的容量提高了100倍。

电子电路是由电子的流动驱动的，它们拥有可测量可控制的电荷。但是，电子不仅流动，而且像微小的磁铁那样会旋转。依据它们的取向，旋转被说成“上”或者“下”。这个额外的变量，或“自由度”，意味着电子可以做比常规电子电路中更多的事和传送更多的信息。

最直接的研究目标就是生产磁随机存取存储器(MRAM)，它利用磁学原理而不是电荷来储存数据。与PC机中的动态RAM不同，MRAM是非易失的。

IBM正在与慕尼黑的Infineon技术公司合作，据称，最早在2005年就能生产出MRAM。它比DRAM快50倍，比静态RAM的密度高10倍，最终它能替代这两种存储器。

还有人认为，MRAM可能替代磁盘做数据存储。将逻辑电路和存储放在同一芯片中，能消除慢速的磁盘I/O，这可是多数计算机处理中的瓶颈。

IBM的MRAM利用了磁隧道结，它应用了旋转电子学，其中电子被允许“隧道”穿过两层基于旋转的铁磁层。每个结能储存一位。它有望成为一种极高性能的通用RAM，即高的读写速度加上极高的密度和非易失性。

研究人员还在进一步开发旋转的更神奇应用。加州大学圣巴巴拉分校旋转电子学和量子计算中心主任David Awschalom正在研究利用原子核的旋转能做些什么。

他说：“原子的亚原子部分(即原子核——译者注)存储信息，而电子起到运送信息进出原子核子系统的作用。”

他的目标是制造基于光学的信息处理器，其中光束通过电子向原子核传送信息。他说，这样的核存储器比传统的半导体存储器在密度和速度上要高出很多量级。

实际上从更广义的角度，旋转电子学研究的迅猛进展将会在半导体材料中把电子学和光子学与磁学结合起来，而传统上这些学科只涉及金属。这将实现极快的、极高效的亚微米器件，将计算、通信和存储结合在一起。不同材料之间的慢速界面(将一种信号或特性转换成另一种)将一去不返，通常会使数据从一个处理阶段过渡到另一阶段步伐放慢的反应时间也将大大缩短。

美国国防高级研究计划局(DARPA)的Wolf说:“你将所有的东西都集成在简单得多的电路中，它们除了电流是旋转-极化的外，很像现在的半导体器件。”例如，它们可以搭建极快速的通信交换机。他说：“你可以把它叫做旋转光电学。它们很容易工作在太赫兹的速度。”

半导体器件不能利用旋转，除非找到办法把旋转极化电子放进去，业已证明这是一件困难的事。但是最近IBM演示了能利用磁隧道结将电流注入进去，如同他们在为MRAM所做的那样。

IBM的Parkin称，旋转电子学半导体可以用于制造可配置的逻辑器件。他说:“通过飞快地重新安排(逻辑)门电路连接的方式，计算机有可能针对某些指令进行优化。”

另一个严峻的挑战是生成能在室温下保持旋转状态的磁半导体。但在今年，物理学家、材料科学家和工程师们在这方面取得了长足的进步。旋转电子学的快速发展有可能继续下去，理论已经很完善，尽管还有很多挑战。