摩尔定律50岁了，它将很快寿终正寝吗？

    1950年时，全球只有不到10台数字计算机。时年33岁的新泽西贝尔实验室科学家比尔•芬尼在这一年发现了一种提纯锗、硅等元素的方法。他当时可能想不到，这一发现促成了硅制微芯片的诞生，推动了计算机和互联网的发展，导致了信息时代的出现。如今，全球已有超过100亿台联网设备。而所有这些设备中，都至少有一块这样的微芯片充当“引擎”作用。

    50年前，一个预言式的定律巧妙地揭示了科技不断进步背后的原因，它就是摩尔定律。微芯片上有许多由纯净硅制成的微型电子开关，它们被称作晶体管。而摩尔定律认为芯片上的晶体管数量每年都会翻倍。1975年，戈登•摩尔修正了他的预测，认为晶体管数量会每两年翻一倍。从此以后，这一定律从未失准。

    为何摩尔定律经年不衰？因为晶体管数量的倍增让计算机芯片能够搭载越来越复杂的电路系统，它们不仅节能，还十分便宜。这导致了计算机、手机的普及，推动了信息技术的革命。

    计算机的价格比起40年前，已然便宜了一千万倍，而一部智能手机拥有的计算能力，已经超过了20世纪90年代计算机科学家使用的工作站。至今为止，摩尔定律依然适用，也因为如此，信息的电子流通变得商品化，改变了我们当中许多人学习、储蓄、旅行、沟通和社交的方式。

    以使用手机进行社交为例。之所以能实现这一点，是因为从1980年（当时笔者才刚进入工程学院）至今，晶体管的价格降低了几百万倍，计算的能效提高了几万倍。因此，售价200美元、由一块饼干大小的电池作为能源的智能手机中，拥有一块包含几十亿晶体管的微芯片，其计算能力足以对图片进行数字化加工，运用强大的数学运算能力编码其数据，再通过无线网络上传和分享它。这就是摩尔定律作用下的成果。

    然而，在它诞生50周年之际，已有迹象表明摩尔定律下的增长开始放缓，我们也几乎可以确定，在接下来的十年内，它恐怕将不复成立。硅晶体管继续微型化下去，将达到仅含有少量原子排列的维度，根据物理定律，这种情况下晶体管和电子电路将无法有效工作。随着摩尔定律下的增长放缓，其他领域的创新，如软件方面的发展，将在短期内补上这一缺口。

    但从长期来看，从20世纪50年代至今未曾改变的传统计算机的基本设计，将会出现根本性的变革。如今的计算机可用来进行精确计算，但它们无法高效地从大量数据中得出推断，做出定性决策或识别模式。下一个实质性的飞越将会出现在那些拥有类人认知能力且高能效的计算机上。IBM的计算机系统沃森在2011年的智力挑战节目“Jeopardy!”上取得胜利，但它消耗的能量是与它同台对垒的人类选手的4000倍。这一经历凸显了对新型高能效计算机的需求。它们要与使用顺序计算方法的经典计算机有所区别，设计者也许将从生物大脑的运转方式中汲取灵感。

    一位记者最近问我，摩尔定律的持续是不是不可或缺的。人类创新这项集体活动之美，就在于保证了没有什么对于技术进步而言是不可或缺的。几十年后，人们或许会将摩尔定律的年代看作黄金时代，这个年代的计算机起初是一个行业实力的体现，后来计算机逐渐小型化，全球的工厂制造了数以亿计一模一样的完美微型电路。但就像候鸟群会以V字队列盘旋在领头者旁边一样，未来会有许多其他技术引领着我们在信息时代继续前进。（财富中文网）

    本文作者为IBM公司物理科学部门总监。

    译者：严匡正

    In 1950, at a time when there were fewer than 10 digital computers worldwide, Bill Pfann, a 33-year-old scientist at Bell Laboratories in New Jersey, discovered a method that could be used to purify elements, such as germanium and silicon. He could not possibly have imagined then that this discovery would enable the silicon micro-chip and the rise of the computer industry, the Internet, and the emergence of the information age. Today, there are about 10 billion Internet-connected devices in the world, such as laptops and mobile phones, and at the heart of each of these devices, there is at least one such micro-chip that acts as its “engine”.

    The reason behind this relentless progress is neatly contained in a prophetic law that was announced 50 years ago this Sunday, called Moore’s Law. The micro-chip is built with tiny electrical switches made of purified silicon called transistors and the law stated that the number of transistors on a chip would double every year. In 1975, Gordon Moore revised his forecast to state that the count would double every two years. The law has held true since.

    Why is Moore’s Law relevant? Because this doubling of the number of transistors led to computer chips that could be packed with increasingly sophisticated circuitry that was both energy efficient and cheap. This led to the widespread adoption of computers, mobile phones, and the information technology revolution.

    The price of computation is about 10 million times cheaper than it was 40 years ago, and the computing power held in a smart phone outstrips the workstations that computer scientists used in their offices in the 1990s. That we have been able to so far hold true to Moore’s Law is the reason that the electronic circulation of information has been commoditized, changing the way many of us learn, bank, travel, communicate and socialize.

    Take the example of social networking using a mobile phone. It works because the cost of a transistor has dropped a million fold and computing is about 10,000 times more energy efficient since 1980, when this writer first went to engineering school. Consequently, a $200 smart phone powered by a biscuit-sized battery contains a micro-chip with a few billion transistors in it and enough computing power to digitally process an image, and then upload and share it wirelessly using powerful mathematics to encode the data. This is a consequence of Moore’s Law in action.

    Yet, on its 50th anniversary, there are tell-tale signs that Moore’s Law is slowing, and we are almost certain that the law will cease to hold within a decade. With further miniaturization silicon transistors will attain dimensions of the order of only a handful of atoms and the laws of physics dictate that the transistors and electronic circuits will cease to work efficiently at that point. As Moore’s Law’s slows down, innovations in other areas, such as developments in software, will pick up the slack in the short-term.

    But in the longer-term, there will be fundamental changes in the essential design of the classical computer that, remarkably, has remained unchanged since the 1950s. Designed for precise calculations, today’s computing machines do not make inferences, and qualitative decisions, or recognize patterns from large amounts of data efficiently. The next substantive leap forward will be in computers with human-like cognitive capabilities that are also energy efficient. IBM’s Watson, the computing system that won the television game show Jeopardy! in 2011, consumed about 4000 times more energy than its human competitors. This experience reinforced the need for new energy efficient computing machines that are designed differently from the sequential, calculative methodology of classical computers and are inspired, perhaps, by the way biological brains work.

    A journalist recently asked me whether the continuation of Moore’s Law was indispensable. It is the beauty of the collective enterprise of human innovation that which ensures that nothing is indispensable indefinitely for technology to progress. Decades later one might look at the era of Moore’s Law as a golden period where computers came of age through a masterful display of an industry’s ability to miniaturize and create billions of flawless and identical copies of tiny circuits at factories throughout the world. But, much as a pack of migratory birds flying in V-formation rotate in at the lead position, there will, at that future time, be many other technologies that will have carried us forward in the information age.

    Supratik Guha is director of physical sciences at IBM.