半导体行业：洗牌1

via 《The Economist》《经济学人》中文版 by 经济学人 on 4/11/09

[2009.04.02]Under new managemen 半导体行业处于洗牌中

The semiconductor industry
半导体行业

Under new management
处于洗牌中
Apr 2nd 2009 | DRESDEN
From The Economist print edition

Chipmakers were suffering even before the global economic downturn. Recession is heightening the pain and highlighting changes in structure and ownership
在全球经济低迷之前，芯片制造业者已经处于危机中。经济危机的到来更加加剧了这种痛苦，并使得芯片制造业在产业结构和所有制方面的转变迫在眉睫。

MOST tourists come to Dresden to view the city's architectural wonders. Beautifully rebuilt, the Frauenkirche (Church of Our Lady), for instance, reveals no hint that its huge cupola once crumbled after a rain of British bombs. But the capital of the German state of Saxony also has more contemporary attractions—at least for technically inclined travellers. It is the hub of one of Europe's biggest technology clusters. Silicon Saxony, as the region has come to be called, boasts 1,500 high-tech firms employing 43,000 people, most of them in the semiconductor industry.

大多数游客来到德累斯顿（Dresden），是为了观赏城市中的建筑奇迹。譬如，Frauenkirche （圣母教堂），它在战后被完美的重建了，以至于没有留下巨大圆顶阁楼曾被英军的狂轰滥炸摧毁过的蛛丝马迹。但是德国萨克森州的首府[即德累斯顿]也同样不乏当代的名胜——至少对于爱好技术的游客们来说如此。事实上，它是欧洲最大科学技术密集区的枢纽—1500高科技企业以及他们的43000员工聚集此处，其中大多数从事半导体行业，也因此，德累斯顿被尊称为硅-萨克森（Silicon Saxony）。

Yet industrial tourists had better hurry. Recently Silicon Saxony has taken some hits that have weakened its foundations. On April 1st Qimonda, a maker of memory chips and the cluster's largest employer, mothballed its factory, having been forced into insolvency earlier this year. Its last hope is to be bought by an outside investor lured by money from the Saxon government. Inspur, a Chinese computer-maker, is among those expressing interest in Qimonda, which has developed some cutting-edge technology.

然而工业旅游者得抓紧时间了。硅-萨克森最近受到的一些打击已经消弱了他的地位。四月一日，地区最大雇主，存储芯片制造厂商奇梦达（Qimonda）关闭了他的工厂，在今年早些，它已经被迫破产。它的最后希望是被外部投资者收购，因为来自萨克森政府的补贴金仍然对投资者具有吸引力。众多投资者表达了对已经发展了一定先进技术的奇梦达的兴趣，其中包括中国电脑制造商浪潮集团（Inspur）。

At Dresden's other big "fab", as chip-fabrication plants are called, is an indicator of another change that may prove just as damaging. There is a new logo at the entrance: visitors are no longer welcomed to AMD but to Globalfoundries. AMD, a maker of microprocessors for personal computers (PCs), decided last year to spin off its fabs into a separate company and to sell a majority stake to investment funds controlled by the government of Abu Dhabi. A good deal of production, some fret, may eventually move from Dresden to the Gulf.

在德累斯顿的另外一个被称为芯片制造工厂的大晶圆厂则表现了一种纯粹破坏性的的改变。在它的入口处有一个新LOGO：欢迎客户来到环球晶圆代工（Globalfoundries）而不是AMD。个人PC微处理器制造商AMD去年决定把晶圆厂分拆到单独的公司，并把多数股份卖给阿布扎比（Abu Dhabi）政府把持的投资基金。有些人担心大量的生产可能将最终从德累斯顿转移到海湾地区。

The likely death of Qimonda and the birth of Globalfoundries* have turned Silicon Saxony into an industrial showcase of a very different kind. It is a visible token of how hard recession around the world has hit the semiconductor industry, which had already been weakened by one of its periodic downturns. Just as important, it demonstrates the longer-term upheavals in the industry. The semiconductor business is becoming less vertically integrated and more concentrated. And its centre of gravity is shifting eastwards.

奇梦达的临近消亡和环球晶圆代工厂的出现已经置硅-萨克森于一种工业展示的奇异境地。而这正是全球经济不景气打击本已受到周期性低迷影响的半导体行业的明显标志。同样重要的，它也是行业长期巨变的明证。半导体业务正从纵向一体化向更多的集聚转变，并且它的重心正向东方转移。

Despite a few signs that the worst may be over—Asian chipmakers' share prices soared recently after shortages were predicted—the industry is still in the midst of the longest slump in its 50-year history. If market researchers are right, it will shrink again in 2009 before resuming growth in 2010. iSuppli, one such forecaster, thinks that revenues will fall by more than 20% this year, to $205 billion (see chart 1). Other observers have been making similarly gloomy predictions.

在普遍预言萧条之时，亚洲芯片制造商近期股价飞涨，这成为少有的显示最坏时期可能结束的标志；除此不论，整个行业仍处于其50年历史中持续最久的低谷之中。如果市场调研正确，在2010年回复增长之前的2009年，半导体行业将会再次收缩。持有如上预测的iSuppli[1]公司认为今年的相关收入将会同比下降20%，为2050亿美元（见图一）。其它观察家也有相似悲观预见。

To understand why the semiconductor industry has been so pummelled, think of integrated circuits (ie, chips) not as tiny pieces of silicon engraved with millions of transistors, but as an essential resource. Before long every man-made object will come with at least one embedded microchip (see chart 2). Jerry Sanders, AMD's founder, once called chips "the crude oil of industry". This seems apt: integrated circuits have become the grease of the information economy. The flip side is that chipmakers have come to depend increasingly on the health of the rest of the economy.

为了解半导体行业为何受到如此沉重打击，应想到集成电路（即芯片）并不仅仅是用成百万晶体管雕刻的小硅片，而更是一种基础资源。不久之后，任何人造物都将被植入至少一个微芯片（见图二）。AMD创始人杰里•桑德斯[2]（Jerry Sanders）曾经称芯片为"工业的原油"。这似乎容易想到：集成电路已经变成信息经济的润滑剂。另一面，芯片制造商也日益依赖于其它经济方面的健康状态。

The chip cycle
芯片周期
However, the industry's own economics are also to blame. Even without the world's wider troubles, these would have caused problems. In explaining how, Dan Hutcheson, chief executive of VLSI Research, a consultancy, likens semiconductor manufacturing to a different industry: farming. Investment decisions have to be made long before products can be sold. Chip farmers have to spend billions and wait years before they can start etching circuits onto "wafers", those thin disks of semiconductor material, the size of pizzas, which are sliced into hundreds of chips at the end of the production process.

然而，行业自身的经济情况也应受到指责，即使没有全球性的危机，这也会导致问题产生。为了解释这个问题，作为顾问的VLSI总裁Dan Hutcheson把半导体制造业比作另一个行业—农业。远在产品能销售之前，投资决定已经做出。芯片农夫不得不花数十亿美元并等待数年直到能开始在晶圆片上雕刻电路。晶圆是由半导体材料制成的披萨饼尺寸超薄原片，在生产流程的最后阶段被切割成成百的芯片。

This goes a long way towards explaining why chipmakers, like farmers, have a tendency to oversupply the market, particularly if they sell memory chips, an undifferentiated product (like winter wheat). Even if prices fall below costs, they have an interest in keeping their fabs humming, in order not to lose their heavy upfront investment and to recover the variable costs. What is more, they are caught on a "technology treadmill", in the words of Mr Hutcheson. Competition forces them always to employ the latest technology, which both increases output and puts pressure on prices.

这里走了一大段长路来解释为什么芯片制造者像农夫般有过度供给市场的倾向，同时这种倾向在他们出售电脑芯片这种无差别产品（类似冬小麦）时更为明显。为了避免损失前面的大宗投资并回收可变成本，即使价格低于成本，他们也十分在意他们的晶圆厂处在继续运行中。更重要的是，用Hutcheson先生的话说，他们被"技术跑步机"俘虏了。竞争促使他们总是去采用最先进的技术，这在增加产出的同时，也加大了价格上的压力。

Finally, just as in agriculture, governments further fuel this innate tendency to oversupply. In prestige, national security, industrial policy or just a desire to create jobs, politicians have always found a reason to support their semiconductor industries, mostly with cash. Silicon Saxony, for instance, has received more than euro1.5 billion (nearly $2 billion at today's exchange rate) from the state of Saxony alone, much of it to coax AMD into investing.

最后，如同农业，政府进一步推动了过度共给的固有趋势。虽然大体上是为了钱，但政客总能找到相当的理由来支持半导体工业，如声望，国家安全，工业政策或仅仅是促进就业。譬如，硅-萨克森单从萨克森州就已经得到超过15亿欧元（当今汇率下约等20亿美元）的财政支持，其中很大一部分用于劝诱AMD继续投资。

Asian governments have been the most active. Thanks to Taiwan's industrial policy, more than half of the world's chips are now made there. Support from the South Korean government made Samsung and Hynix the world's biggest makers of memory chips; they supply about 50% of this segment. China seems intent on turning its semiconductor companies into market leaders at almost any price, above all Semiconductor Manufacturing International Corporation, or SMIC. All this explains why of the 40 fabs under construction in 2007, 35 were in Asia, three in America and only two in Europe.

亚洲政府已经是最活跃的了。由于台湾的工业政策，世界上超过半数的芯片在此制造。同时，在韩国政府的支持下，三星和现代成为世界最大的内存芯片制造商，提供世界该产品总量的50%。中国政府似乎决心付出任何代价将他的半导体公司转变为市场领导者，首先是中芯国际，即SMIC。所有这些解释了为什么2007年兴建的40个晶圆厂有35个在亚洲，3个在美国，只有两个在欧洲。

Not surprisingly, at times supply far outstrips demand. From 2002 until last year Asian makers of memory chips, especially, invested as if capital were free—which explains why everybody is now bleeding money. In July 2007 the price of a DRAM (dynamic random access memory) chip with a capacity of 512 megabits was more than $2. In early April it was about 50 cents. Smaller makers cannot cope. Qimonda, for instance, piled up losses of about euro1.5 billion between October 2007 and June 2008. Its revenues were only euro1.3 billion.

不必惊讶，有时供给远大于需求。从2002年到去年，尤其是亚洲内存芯片制造商，好似不要钱般的投资，这也解释了为什么现今每个人都大赔特赔。2007年 7月512MB的DRAM（动态随机存取存储器）芯片的价格超过两美元，今年4月早些降至50美分。小型制造厂商无法应对，例如奇梦达，它在2007年 10月到2008年7月之间累积了15亿欧元的亏损，而他的收入仅为13亿欧元。

Given the scale of the losses and the screaming from other industries, governments look less inclined to help this time. Even Taiwan is having second thoughts about an ambitious plan to save its memory-chip industry, announced only last month. The idea is to merge and bail out the country's six makers of memory chips, which have lost $12.5 billion in the past two years and accumulated $11 billion in debts.

尽管看到亏损的规模和其他行业的呼吁，政府似乎并不倾向于此时出手。即使台湾也正考虑改变上月刚公布的关于解救内存芯片产业的庞大计划。那个计划是合并本土的六家内存芯片制造商以摆脱困境。这些制造商在过去的两年里亏损了125亿美元并积累了110亿美元的债务。

Even if Taiwan were to let these firms fail, which is highly unlikely, supply would still exceed demand, according to iSuppli. Global sales of memory chips will not start growing again before next year. And growth will not reattain its 2006 rate before 2015.

来自iSuppli的数据指出，这些企业关闭是不太可能的；但即使台湾关闭这些企业，供给仍然大于需求。全球内存芯片销售在明年之前不会开始增长，此外，增长速率在2015年之前不会重新达到2006年水平。

Whatever happens to Qimonda and its Taiwanese rivals, the current crisis is sure to speed up two seemingly contradictory long-term trends in the industry. It is consolidating, in that the manufacture of chips is becoming concentrated among fewer companies. At the same time, it is splitting up, in that more companies are specialising in design, and contracting out or quitting the making of chips. Both developments are mainly the consequence of what has come to be called "Moore's Second Law", an economic counterpart to a better known observation by Gordon Moore, one of the founders of Intel, the world's biggest chipmaker by revenue.

无论在奇梦达和它的台湾对手身上发生了什么，当前的危机必然会加速该行业这两个看似矛盾的长期趋势。趋势之一是半导体行业的整合—芯片制造集中到少数的企业。同时，另一个趋势是行业的分化，更多企业专注于设计，外包或退出芯片制造业务。这两种趋势主要都是人们所说的"摩尔第二定律"的结果，该定律是戈登 •摩尔提出的另一更著名论断在经济学上的应用。这是Intel的创始人之一戈登•摩尔[3]（Gordon Moore）提出的一个著名经济学论断，就收入而言，Intel是世界上最大的芯片制造商。（这里说的另一更著名的论断便是"摩尔定律"——在价格恒定的情况下，集成电路上可容纳的晶体管数目，约每隔18个月便会增加一倍，性能也将提升一倍；或者说，每一美元所能买到的电脑性能，将每隔18个月翻两倍以上。）

The original Moore's Law is usually summarised thus: the number of transistors on a chip doubles every 18 months. In fact Mr Moore first predicted this would happen every year and later changed his forecast to every two years; the average has become his law. Mr Hutcheson points out that Mr Moore made more than a purely technical prediction. He also stated that the cost of an integrated circuit would stay the same, a halving of the cost per transistor with every doubling of the number.

原始摩尔定律经常被表述为：单块硅芯片上所集成的晶体管数目每十八个月增加1倍。事实上，摩尔先生最初的的预测为每一年，后来改为每两年；他的这条定律取其平均数。Hutcheson先生指出摩尔先生做的不仅仅是纯粹的技术预测。他接着指出由于每个晶体管的成本减半和能集成的晶体管数量翻一番，集成电路的成本将保持不变。

This has turned out to be essentially correct, but progress has come at a high price. The ever more sophisticated equipment required to make semiconductors has been getting dearer with every iteration of Moore's Law. The most advanced chips are built using 32-nanometre technology, meaning that transistors are now so tiny that more than 4m can fit on this full stop. Lithographic tools for transferring Lilliputian circuitry onto a wafer cost up to $50m a pop. To reach the economies of scale needed to make such investments pay, chipmakers must build bigger fabs.

摩尔定律已经被事实证明是十分正确的，但进展是在高昂代价下的。随着摩尔定律的每个反复，用来制造半导体的越来越尖端的设备已变的更为昂贵。最先进的芯片使用32纳米技术制造，这意味着晶体管是如此微小，以致于超过4百万个才抵得上这个句点。流行的用来移动微小电路到晶圆上的印刷工具价值5千万美元。为了达到规模经济以抵消大量的前期投入，芯片制造商必须建造更大的晶圆厂。

Rising fixed costs give rise to Moore's Second Law: as the cost of transistors comes down, the cost of fabs goes up, albeit not at quite the same rate. In 1966 a new fab cost $14m. By 1995 the price had risen to $1.5 billion. Today, says Intel, the cost of a leading-edge fab exceeds $6 billion, including all the preparatory work. And the Taiwanese Semiconductor Manufacturing Company (TSMC) has built two "GigaFabs" for between $8 billion and $10 billion each, which would buy you four nuclear power stations. The output of such monsters depends on the mix of products, but they each could easily churn out 3 billion chips a year.

固定成本上升引起摩尔第二定律：尽管不是完全相同的速率，随着晶体管成本下降，晶圆成本上升。在1966年，建设一个新晶圆厂的花费为1400万美元，到了1955年花费上升到15亿美元。今日，就Intel而言，其先进晶圆厂的建设花费超过60亿美元，其中包括所有准备工作。并且，台积电（TSMC）已经兴建了两个"十亿级晶圆厂"，每个的造价都在80亿到100亿美元之间，这些钱足够买四座核电站。这些怪物般的产出有赖于产品的混合，但他们中的每一个都能一年轻松产出30亿芯片。

These ever-increasing costs and the need for specialisation have caused the industry to splinter, says Derek Lidow, iSuppli's chief executive. Originally, all chipmakers were vertically integrated, meaning they designed the chip, built the tools to make them, ran the fabs and added the necessary connectors. As costs went up and certain activities became more and more complex, they were spun out to spread expenses and know-how. Semiconductor equipment, design software and packaging have long been done by separate companies. But the past ten years have seen the rise of "fabless" firms, which merely design integrated circuits.

iSuppli总裁Derek Lidow指出，这些成本的不断增加和对专业化的需求已经导致整个行业开始分化。最初，所有的芯片制造者呈纵向一体化，意味着他们自己设计芯片，制造制作工具，建设晶圆厂以及添加必要的中间环节。随着成本的增加和相关活动的复杂化，他们开始分化以分摊开支和工艺流程。半导体设备、设计软件程序包已长期分别由单独的公司来完成。过去的十年里已经出现了许多"无生产线"公司，他们仅仅设计集成电路。

Now established chipmakers can no longer afford to develop their own manufacturing processes or even to run their own fabs. To share the pain, IBM, Samsung and others have teamed up to use chipmaking technology jointly. Some firms, such as Texas Instruments, have chosen to go "fab-lite", meaning that they have their own fabs only for certain chips. Others, such as AMD, have spun off manufacturing completely (although AMD's decision had much to do with a lack of cash after it bought ATI, a maker of graphics chips, for $5.4 billion in 2006).

现在已建立的芯片制造商再也不能发展自己的生产工艺或者运行他们自己的晶圆厂。为了分担痛楚，IBM、三星和其它公司已经联合起来共同使用芯片制造技术。一些公司，例如德州仪器（Texas Instruments），已经选择走"轻晶圆厂"道路，也就是说，他们有自己的只生产特定芯片的晶圆厂。其它的，例如AMD，它已经把制造完全的分割了出去（虽然AMD的决定在很大程度上是因为在2006年以54亿美元的价格收购了图形芯片制造商ATI后缺乏资金）。

Hence the rise of "foundries", the smelters of the information age. These are essentially contract manufacturers. Although far from household names, they are huge companies, churning out about one quarter of the world's semiconductors. The biggest, TSMC, has a manufacturing capacity greater even than Intel's. Its revenues grew at an annual average rate of 13% for several years, topping $10.6 billion, before falling by almost a third in the last quarter of 2008.

因此晶圆代工厂的增加是信息时代的熔炉。这些代工厂基本上属于合同制造商。虽然远非家喻户晓，但他们都是巨型企业，以致制造全世界半导体的四分之一。其中最大的—台积电，它的制造能力甚至超过Intel。它的年收入以13%的年增长率增长，在2008最后季度几乎三分之一的下降之前，曾达到过106亿美元的高点。

TSMC also illustrates a corollary of Moore's Second Law: even the biggest chipmakers must keep expanding. Intel today accounts for 82% of global microprocessor revenue and has annual revenues of $37.6 billion because it understood this long ago. In the early 1980s, when Intel was a $700m company—pretty big for the time—Andy Grove, once Intel's boss, notorious for his paranoia, was not satisfied. "He would run around and tell everybody that we have to get to $1 billion," recalls Andy Bryant, the firm's chief administrative officer. "He knew that you had to have a certain size to stay in business."

台积电还表明了摩尔第二定律的必然结果：即使最大的芯片制造商也必须保持扩张。Intel之所以现今占据全球微处理器销售收入的82%，年收入达到376亿美元，就是因为它早已懂得了这个道理。早在20世纪80年代，Intel的市值为7亿美元，这在当时已相当庞大了。而当时的因偏执而声名狼藉的老总Andy Grove并不满意。"他总是东奔西跑并告诉每一个人我们应当达到10亿美元，" 该公司的首席行政长官Andy Bryant回忆道。"他知道，你必须有一定的规模来维持生意。"