英特尔坚称摩尔定律并未失效

为了改善前景，英特尔通常会推介自己销售的产品。但上周二，这家芯片巨擘一反常态，深入地剖析起自己的制造技术来。

英特尔这样做旨在证明，尽管推出更小芯片的速度已经放慢，但摩尔定律依然有效——英特尔联合创始人戈登•摩尔预言，芯片的晶体管密度至少每两年就会翻一番。此举的另一个目的是为自己不断加大投入的芯片代工业务吸引更多客户。

上周二，英特尔高层并未谈论该公司可能怎样支持下一波镜面笔记本电脑、自动无人机和自动驾驶汽车。相反，他们解释了英特尔怎样把更多的晶体管和其他特性集成到晶圆中，从而造出更强大并且更省电的芯片。显然，他们的意图之一就是重置那些围绕在摩尔定律四周的预期——由于近年来未能坚持此前几十年的惯例，也就是每两年就把晶体管的尺寸压缩一次，英特尔已经遭到了投资者和分析师的惩罚。

过去五年中，英特尔的股价仅上升了28%，还不到标普500指数涨幅的一半。出现这样的差距绝不只是因为英特尔的业绩没有达到预期，而是因为它错过了移动化浪潮。个人电脑及其使用的英特尔芯片的销售都受到了这股浪潮的沉重打击。

几十年来，英特尔一直奉行钟摆（tick, tock）战略。在工艺（tick）年，该公司改良芯片的途径是加大晶体管密度和提高制程精细度，比如从32纳米工艺转向22纳米工艺。在之后的架构（tock）年，制程工艺将保持不变，英特尔则会通过加入新的特性来提高芯片的性能。

但英特尔并不总是能按两年一次的节奏来提高精细度。该公司的制程工艺从45纳米变为32纳米花了大约27个月，从32纳米到22纳米用了28个月，从22纳米到目前的14纳米则用了30个月。同时，从2014年9月开始，英特尔在制程工艺方面就再无进展。新的10纳米芯片预计最终会在2017年底问世。

上周二的情况介绍旨在阐述英特尔用于改良芯片的所有其他技术，并且放大这个聚焦点。

该公司负责制造业务的执行副总裁斯泰西•史密斯对《财富》杂志表示，“这个世界不断向前迈进”，英特尔则“一直在寻找那个优化点”，以便利用各种各样的方式让自己的芯片每年都得到提升。他说：“我们对消费者的承诺是我们每年都这么做一次。”

在活动中，总裁穆尔西•兰德奇塔拉负责解释英特尔的战略调整。他说：“我们准备放弃钟摆战略，并用一种基于创新风潮的战略取而代之。”兰德奇塔拉指出，随着英特尔开始采用新工艺来制造芯片，随后几年该公司每年都会有“一波”改进。

这次活动持续了四个多小时，它的另一个目的是吸引一个不断扩大的群体，那就是能设计芯片但没有半导体工厂的公司。以苹果公司为例，iPhone 7用的就是苹果自行设计的A10处理器，但它由别人制造，只不过生产方并非英特尔。

首席执行官科再奇一直要把英特尔塑造成供租用的芯片制造商，也就是用自己的先进设备制造其他公司设计的芯片，而不是只生产自行设计的产品。

经过大规模整合，半导体行业中有能力制造最先进芯片的公司已从10年前的13家变为如今屈指可数的几家。除了英特尔，目前这份名单上只剩下三星、台积电和2009年从AMD拆分出来的格罗方德。

执行副总裁史密斯说：“我们不仅仅是愿意这样做，我们甘之如饴。”他还说周二的活动“有点儿像新人联谊会，我们一直在建设这些产能，而且在此期间发布了一些消息……今天我们把所有这些和盘托出。”（财富中文网）

译者：Charlie

审稿：夏林

Intel's efforts to improve its prospects usually involve pitching the products it sells, but on Tuesday the chip giant went in another direction, offering a deep dive into the company's manufacturing techniques.

The goal was to prove that Moore's Law—the prediction by the Intel co-founder that transistor density can double every two years or less—remains in force, even though the rate of introducing smaller-scale chips has slowed. A second aim was to attract more customers in Intel's growing effort to manufacture chips for others.

Instead of talking about how it might power the next wave of slick laptops, automated drones, or self-driving cars, Intel executives on Tuesday explained how the company could make its chips more powerful and energy efficient by cramming more transistors and other features onto the silicon wafers. One clear aim was to reset expectations around Moore's Law, as the company has been penalized by investors and analysts in recent years for failing to meet its decades-old strategy of shrinking the size of transistors every other year.

In the past five years, Intel's stock price has gained only 28%, less than half the gain of the S&P 500. But this lag is about far more than missed expectations, as the company missed the mobile wave, which has greatly hurt sales of PCs and the Intel processors that run them.

For decades, Intel (intc, +0.59%) followed what it called a "tick, tock" strategy. In one year, representing the "tick," it would improve its microprocessors by printing the transistors on the chips closer together, reducing the scale of the process, say, from 32 nanometers to 22 nanometers. In the second year, the "tock," the scale would stay the same and Intel would improve the chips by adding new features.

But Intel hasn't been able to hit the every-other-year schedule for reducing scale. To get from 45 nm to 32 nm took about 27 months, 28 months to go down from there to 22 nm and 30 months to shrink to the current 14 nm process. And that's where Intel has been stuck since September 2014. New 10 nm chips are finally expected towards the end of 2017.

The point of Tuesday's presentations was to explain—and magnify the focus on—all the other techniques Intel uses to improve its chips.

"The world moves on," Intel executive vice president Stacy Smith, who oversees the manufacturing side, tells Fortune. The company is "constantly looking to find that optimization point" of all the different methods for improving its chips every year. "And our commitment to our customers is that we do that on an annual cadence," he says.

It fell to Intel president Murthy Renduchintala during the presentations to explain the changing metaphor. "We're going to do away with the tick-tock metaphor and replace it with a metaphor based on waves of innovation," he said. As Intel moved to a new scale for printing chips, it could would follow for several years with annual "waves" of improvements, he explained.

Another goal of the over-four-hour program was to attract the growing legions of companies that design chips but don't own their own semiconductor manufacturing factories, for example, runs the iPhone 7 using its custom A10 chip which is made by others, though not Intel.

CEO Brian Krzanich has been building Intel's business as a chip maker for hire, printing up chips designed by other companies at its cutting-edge foundries, instead of using the factories exclusively for its own designs.

With massive consolidation in the semiconductor industry, only a handful of companies can even compete to make the most advanced chips, down from 13 ten years ago. Beyond Intel, the list currently includes only Samsung, Taiwan Semiconductor and GlobalFoundries, the AMD spinoff created in 2009.

"It's something we're not just willing to do—we're embracing it," Smith says. Tuesday's presentations were "a bit of a coming out party," he says. "We've been building these capabilities, we've had some announcements along the way... but today we put it all together."