Moore’s Law Running Out of Room, Tech Looks for a Successor
By JOHN MARKOFF

SAN FRANCISCO — For decades, the computer industry has been guided by a faith that engineers would always find a way to make the components on computer chips smaller, faster and cheaper.

But a decision by a global alliance of chip makers to back away from reliance on Moore’s Law, a principle that has guided tech companies from the giant mainframes of the 1960s to today’s smartphones, shows that the industry may need to rethink the central tenet of Silicon Valley’s innovation ethos.

Chip scientists are nearly at the point where they are manipulating material as small as atoms. When they hit that mark within the next five years or so, they may bump into the boundaries of how tiny semiconductors can become. After that, they may have to look for alternatives to silicon, which is used to make computer chips, or new design ideas in order to make computers more powerful.

It is hard to overstate the importance of Moore’s Law to the entire world. Despite its official sound, it is not actually a scientific rule like Newton’s laws of motion. Instead, it describes the pace of change in a manufacturing process that has made computers exponentially more affordable.

In 1965, the Intel co-founder Gordon Moore first observed that the number of components that could be etched onto the surface of a silicon wafer was doubling at regular intervals and would do so for the foreseeable future.

When Dr. Moore made his observation, the densest memory chips stored only about 1,000 bits of information. Today’s densest memory chips have roughly 20 billion transistors. To put it another way, the iPad 2, which went on the market in 2011 for $400 and fits in your lap, had more computing power than the world’s most powerful supercomputer in the 1980s, a device called the Cray 2 that was about the size of an industrial washing machine and would cost more than $15 million today.

That iPad 2, mind you, is slow compared to newer models.

Without those remarkable improvements, today’s computer industry wouldn’t exist. The vast cloud-computing data centers run by companies like Google and Amazon would be impossibly expensive to build. There would be no smartphones with apps that allow you to order a ride home or get dinner delivered. And scientific breakthroughs like decoding the human genome or teaching machines to listen would not have happened.

Signaling their belief that the best way to forecast the future of computing needs to be changed, the Semiconductor Industry Associations of the United States, Europe, Japan, South Korea and Taiwan will make one final report based on a chip technology forecasting system called the International Technology Roadmap for Semiconductors.

Nearly every big chip maker, including Intel, IBM and Samsung, belongs to the organization, though Intel says it is not participating in the last report.

To replace what the semiconductor industry has done for nearly 25 years, a professional organization called the Institute of Electrical and Electronics Engineers announced on Wednesday that it will a create a new forecasting system, called the International Roadmap for Devices and Systems, that is intended to track a wider range of computer technologies.

One technology could be so-called quantum computing, a cutting-edge reimagining of how computers work that taps quantum physics — a branch of physics that explains how matter and energy interact. Another could be graphene, a form of carbon and an alternative to silicon that could produce smaller and faster transistors that use less power.

“The end of Moore’s Law is what led to this,” said Thomas M. Conte, a Georgia Institute of Technology computer scientist and co-chairman of the effort to draw up a new set of benchmarks to replace the semiconductor reports. “Just relying on the semiconductor industry is no longer enough. We have to shift and punch through some walls and break through some barriers.”

Predicting the end of Moore’s Law has for years been a parlor game in Silicon Valley, and not everyone in the industry believes that what it has come to represent is nearly over. Intel, the world’s largest chip maker, is a notable contrarian and predicts it has the means and know-how to push further into the atomic level.

In a statement on his company’s website last month, Brian Krzanich, Intel’s chief executive, played down concerns. “I have witnessed the advertised death of Moore’s Law no less than four times,” he wrote.

Intel, however, faces its own problems because of a long slump in PC sales and an inability to sell many chips to smartphone makers. Last month, the company announced plans to lay off 12,000 workers — roughly 11 percent of its work force — and take a $1.2 billion charge.

The industry saw signs that Moore’s Law was running out of steam as far back as 2005, when researchers began to worry that computer processors were becoming so hot that they would soon match the surface of the sun in heat output.

But the industry managed to fix the problem by worrying less about increasing speed and instead splitting tasks among many processors. In effect, it kept things cool by finding a way to share the load.

By walking away from a task they have managed for so long, the chip makers other than Intel — the Silicon Valley giant says it is no longer formally contributing to the forecasting process — are posing broader questions about their businesses.

“As you look at Intel saying the PC industry is slowing and seeing the first signs of slowing in mobile computing, people are starting to look for new places to put semiconductors,” said David Kanter, a semiconductor industry analyst at Real World Technologies in San Francisco.

In addition to the impending physical limits of transistors, other barriers are looming. For example, most of the semiconductor industry now insists that the per-transistor cost of computer chips has stopped falling. That had been one of the factors leading to rapid development of new computer technologies.

Many executives and analysts in the computer industry are skeptical about Intel’s ability to keep Moore’s Law going. They point out that if the chip maker were able to continue to reduce costs, it would have been able to make larger inroads into the mobile computing world of smartphones. As part of its recent cutbacks, the company killed its Atom microprocessors, which it had been unsuccessfully trying to sell to smartphone makers.

“If your whole business was about Moore’s Law, and it was ending, how would you react?” Dr. Conte wrote in an email message.

摩爾定律到頭了 半導體業隱憂

指引電腦及半導體產業技術發展數十年的摩爾定律（Moore’s Law）正瀕臨極限，全球工程師原本一直維繫著這個定律，不斷讓電腦晶片更小、更快、更便宜，如今半導體製程終於接近了物理極限，業界將推出新的預測系統預估電腦和半導體業未來發展。

摩爾定律由英特爾創始人之一戈登．摩爾（Gordon Moore）在一九六五年發表。主要指的是當價格不變時，積體電路上可容納的電晶體數目約每隔十八個月就會增加一倍，性能也將提升一倍。之後，摩爾定律還被用於其他行業發展趨勢的分析預測。

但摩爾定律也受限於物理定律，當晶片線寬（產業能製造的最小導線寬度）縮小到廿奈米以下，有人認為摩爾定律將無法繼續預測半導體產業。因為線寬或許能持續在規定時間內縮短，製造成本卻大幅上升。易言之，即便技術上電晶體尺寸能夠循摩爾定律縮小，但商業上成本卻大幅提升至單一公司無法負荷的程度。

摩爾定律長期以來鞭策電腦業做出驚人發展。舉例而言，二○一一年上市售價四百美元的蘋果iPad 2，性能比一九八○年代的超級電腦克雷二號還強，而克雷二號體積如同一台大型工業用洗衣機，造價要一千五百萬美元。

沒有這種發展，今天讓Google、亞馬遜等企業得以運作的雲端資訊中心都將因太貴無法建造，也不會有智慧手機、平板電腦。不過，摩爾定律在二○○五年開始面臨挑戰，當時晶片面臨過熱問題，工程師被迫停止提高時脈、改採多處理器分工方式，繼續維繫該定律。

到了去年，英特爾宣布新一代十奈米製程晶片延緩出貨，紐約時報說，這顯示摩爾定律終於撐不下去了。

國際電機電子工程師學會訂於四日提出新的預測系統，名為「國際設備與系統藍圖」，取代國際半導體業界使用近廿五年的「國際半導體技術藍圖」晶片技術預測系統。

美國、歐洲、日本、台灣、南韓的半導體協會，也將對「國際半導體技術藍圖」提出最後檢討報告。英特爾、IBM、三星等晶片大廠都是成員，但英特爾表示並未參與這次的報告。

半導體的小型化在未來五年將面臨極限。電腦晶片目前是蝕刻出二維導線的矽基板，IBM正研究以奈米碳管和石墨烯當做運算元件的基板，希望這些基板利用的是光脈衝，而不是電壓。

原文參照：
http://www.nytimes.com/2016/05/05/technology/moores-law-running-out-of-room-tech-looks-for-a-successor.html

2016-05-05．聯合報．A15．國際．編譯張佑生