Archives
 
 
 
  Special
 
 
 
  About Us
 
 
 

Newsletter
Free E-mail Newsletter from BYTE.com

 
    
           
Visit the home page Browse the four-year online archive Download platform-neutral CPU/FPU benchmarks Find information for advertisers, authors, vendors, subscribers Request free information on products written about or advertised in BYTE Submit a press release, or scan recent announcements Talk with BYTE's staff and readers about products and technologies

ArticlesHow to Make the World's Fastest CPUs


February 1997 / Features / How to Make the World's Fastest CPUs

Enter a world where dust is your enemy, day and night run together, and blazingly fast CPUs are born.

Tom Thompson

The amber halcyon lights show you right away that Fab 6 is another world. The lights cast an eerie glow that makes everything neither bright nor dull, like a stretch of deserted expressway at 3 a.m. But unlike the real world, the light inside Fab 6 never varies. The natural rhythms of sunrise and sunset go unnoticed, so that once inside this strange environment, you quickly lose track of time.

Fab 6 is housed inside a long, low building at Digital Semiconductor's Hudson, Massachusetts, campus. The fabrication facility is one of the most sophisticated and closely managed manufacturing operations anywhere in the world. It has to be. The main output of Fab 6 is Alpha processors, the fastest general-purpose CPUs now on the market. Recently, the facility has also begun to manufacture StrongARM chips, which power a new generation of hand-held dev ices. These processors have features as small as 0.35 microns, as thin as any commercial processor die on the market today. In this land of minute tolerances, a spec of dust one quarter the thickness of your hair can be your worst enemy.

This fab, like dozens of others throughout the world, is a highly specialized combination of photolithographic facility, chemical plant, assembly line, and testing center. Most remarkable is the level of quality that these facilities produce. Software may crash, hard drives may freeze, but most of us take the reliability of our CPUs for granted. Manufacturing a processor that consists of millions of transistors whose structures are smaller than the wavelength of green light presents formidable challenges. Ye t the manufacturing plants that assemble or fabricate these complex devices produce tens of thousands a month, and at a reasonable cost.

This impressive achievement is possible through a combination of art and science, including scrupulous attention to detail plus the use of complex and expensive manufacturing equipment. To see how this is done, come with us inside Fab 6.

Step 1: Come Clean

Before you enter Fab 6, you must make yourself worthy. First, you walk across mats of tacky adhesive that strip particles from your shoes before you're anywhere near the actual manufacturing processes. Next, you don your bunny suit , the white jumpsuit that traps any dust, skin particles, and lint that may be clinging anywhere from your neck to your toes. A clear shield covers your face and eyes, a soft helmet wraps around your head. For good measure, once you're dressed you must pass through a short tunnel lined with nozzles that blast you with compressed air in a final at tempt to rid you of contamination.

There are reasons for this slavish attention to cleanliness. During fabrication, any contaminants that land on a wafer -- the sheet of silicon that's the foundation for a chip -- can ruin the processors built on it. For example, a dust mote glowing in a beam of sunlight can damage hundreds or thousands of circuits. Even a smoke particle, measuring in at 0.5 microns, can short out a pair of lines in a nascent processor.

Fab 6 isn't entirely contamination free, but it's clean enough for its manufacturing processes to be cost effective. The air is rigorously filtered so that each cubic foot contains no more than one 0.1-micron-size particle. Because ink, paper, and graphite can generate contaminants, Fab 6 is a true paperless office. Everywhere, large, flat-panel plasma screens and networked PCs display the current state of the operation. Workers carry palmtop computers and PDAs for jotting notes about problems encountered during the work day. The staff enters the infor mation into a large on-line database that helps document problems and provides statistical analysis for quality control. Since a system crash could result in a disastrous loss of tracking information, the company uses Open VMS running on a fault-tolerant VAXcluster with disk-shadowing.

Step 2: Meet Your Brothers in Arms

Fab 6 differs from most manufacturing plants because for the most part, machines outnumber people. Throughout the plant, robotic arms pick up, move, and position wafers during the various processing stages. To move wafers in bulk, special racks carry two dozen wafers at a time. An automated materials-handling system transports wafers overhead on a network of rails that shuttles racks of wafers from one part of the facility to another. This machinery minimizes potential defects that could occur due to inadvertent rough handling by humans.

Fab 6 uses silicon wafers 8 inches in diameter, onto which successive layers of chemically-treated, or "dope d," silicon, oxide, and metal are applied (see the sidebar "Building Transistors, Layer by Layer"). These layers assemble the circuits that make up the processor. The fab process continuously repeats several basic operations, which gradually build layer upon layer of material on the wafer. The addition of these layers is an intricate procedure: It can take as few as several weeks to two months for the workers at Fab 6 to apply all the necessary materials. A wafer may ultimately carry dozens of microprocessors on its surface.

Different techniques deposit material on the wafer. A gas-oxidation process stacks layers of silicon and silicon dioxide (a good insulator) over the entire wafer. Sputtering applies metal layers over the wafer. In sputtering, the wafer and a target of the desired metal, such as cobalt, tungsten, or aluminum, are placed in a vacuum chamber. Ions bombard the metal target. Metal atoms dislodged from the target by this bombardment condense on the wafer's surface.

Step 3: Creat e Processor Circuits

However, it isn't enough simply to deposit layers of material on the wafer. To build the patterns that assemble the processor circuits, the fabrication process must be able to selectively apply material onto the wafer. This is accomplished with photolithography, a procedure that photographically transfers patterns onto a surface for etching. This technique is similar to the photolithography traditionally used to etch the plates that print newspapers and magazines. However, to build a processor's microscopic features, the fab's photolithographic operation must be done with great precision and consistency.

A machine applies a thin layer of photosensitive material (commonly called a photoresist) to the wafers. At this point, the wafer might have a layer of metal, silicon oxide, or doped silicon on it. A machine dispenses the photoresist as a liquid onto the wafer. The wafer is then rotated at several thousand RPM to evenly distribute the photoresist across its surface. Next, the wafer is allowed to dry.

A device called a stepper projects the desired patterns onto the wafer. The patterns are called masks because they block areas on the wafer from exposure to light. The masks consist of glass plates with chromium patterns imprinted onto them, and are several times larger than the image projected onto the wafer. Using larger masks makes their defects easy to spot, which in turn reduces pattern defects on the wafer. A special lens reduces the patterns to the desired size.

A robotic arm picks up a wafer and shuttles it into the stepper, which is responsible for making the pattern exposures. It exposes the circuit pattern at one spot on the wafer, the lens steps to a new location, and repeats an exposure at this spot. This methodical stepping process packs as many processor patterns as possible onto one wafer. As mentioned earlier, laying these patterns onto the wafer must be done with great precision. Each processor requires many separate, precisely align ed patterns to build its working elements. The number of exposures varies depending on the complexity of the processor. However, one misaligned or blurred exposure can render the processor's circuits useless.

Fab 6 uses deep ultraviolet light to make the exposures on the photoresist. The funky yellow lighting that illuminates certain areas of the fab serves a purpose: This color's wavelength doesn't carry enough energy to trigger chemical reactions in the photoresist. This in turn simplifies handling of the wafers.

Where the light strikes the wafer, it causes chemical reactions that change the polymers (large molecules) in the photoresist to monomers (small molecules). While the polymer material is insoluble, the monomers are easily removed using solvents.

The monomers are then washed away, leaving a pattern of tough photoresist over other areas. This remaining layer lets the chip designers selectively implant certain materials, or selectively remove (etch) material from the surface. For example, a silicon layer might undergo implantation so that it achieves the required conductive properties, while a metal layer might be etched to remove most of the material except where it makes electrical connections to other circuit elements.

Step 4: Perform Implanting or Etching

Although there are a number of ways to place doped silicon on the wafer, Fab 6 distinguishes itself from many other fab facilities because it uses ion implantation. This technique relies on an implantation unit -- basically a vacuum chamber -- that uses an electric field to accelerate ions of the desired material toward the wafer. When the ions strike the wafer's surface, they become embedded in the silicon, changing its electrical properties. Fab 6 uses arsenic to make negative (or n-type) regions on the wafer, and boron to make positive (or p-type) regions. Ion implantation allows the technicians to control the amount of ions within the doped regions on the wafer. This is done by limiting th e size of the dose (determined by the number of ions launched at the wafer) and how far they penetrate the surface (determined by the voltage intensity).

One advantage of ion implantation is that it can be performed near room temperature. This relatively low temperature prevents fine features already built on the wafer from blurring. By contrast, high-temperature implantation processes can cause fine structure to blur as features melt or diffuse into one another. The downside to ion implantation is that it dislocates the atoms at the wafer's surface. This can be corrected by a short exposure at high temperatures, in a process known as annealing. Ovens heat the wafers to around 1000 C for a few seconds; the wafers are then rapidly cooled. The annealing process allows the surface atoms to recrystallize into their proper orientation, while minimizing the possible detrimental effects of diffusion.

The etching operation removes material from the wafer. At Fab 6 robotic arms carry the wafers in a quartz rac k, which immerses the wafers in an acid bath. The wafers are alternately dunked in baths of hydrochloric acid, hot distilled water, and hydrofluoric acid to remove the desired material. Vertical air flows carry the fumes away from the technicians and equipment.

When implantation or etching is complete, the wafers then move to a rinsing machine that applies a stripping solution to remove the remaining photoresist. The wafers are now ready to undergo a new round of treatments. Another layer of material is applied, and then the deposition photolithography, implantation, and etching operations modify this layer so that it forms a pattern that becomes yet another part of the processor's circuits.

Step 5: Finding Fault

The wafers are periodically inspected for defects that might have occurred during the etching process, and for any residual photoresist. An inspection machine automatically scans the wafers for these problems, and alerts technicians, who use high-powered microscopes to an alyze trouble spots. Information on these defects is entered into the fab's on-line database. Statistical analysis of this data is done constantly to identify potential problems in any part of the manufacturing process.

Once all the layers are applied, the completed wafer is ready to undergo electrical checks. The first test is a parametric analysis, which is an overall assessment of the quality of the work done on the wafer. A testing machine inserts electrical probes onto test circuits placed in the scribe lanes adjacent to each processor. The purpose of the scribe lane is to provide an area where the wafer is cut to free the individual processors. This valuable real estate also serves double-duty by holding the parametric test circuits. The test circuits consist of low-density transistors (there are only hundreds of them occupying an area that normally holds millions of transistors), which undergo a set of electrical tests. If the wafer passes the parametric tests, the overall quality of the wafer's ma nufacture is considered good, and it goes on for more extensive testing. If the wafer fails the test, there's no salvation -- technicians scrap it.

The next set of electrical tests checks the integrity of each processor on the wafer. This is called the probe test, because the test machine places tiny sets of probes onto the processor. One set of probes injects signals into it, and other probes monitor the resulting output signals generated by the processor. A processor that fails these tests is marked by the test apparatus with an ink dot. Subsequent testing usually identifies approximately 10 percent of the processors on the wafer as defective.

Step 6: Assemble Separate Dies

A laser beam slices through the wafer's scribe lanes, carving it up into individual slips of silicon, or dies. Each die has a single processor on it. Defective dies get discarded; viable processor dies move to a chip carrier. Another machine connects aluminum wires that attach the die to the carrier's signal pins. The carrier is sealed in an atmosphere that consists of nitrogen, because this inert gas can reduce the effects of oxidation within the chip carrier.

The processors undergo a final battery of electrical checks that is identical to the probe test. Next, the processors are tested at a variety of clock speeds. Slight variations in the fabrication process can subtly affect the quality of the processor's internal workings. This is why some processors begin to malfunction at higher clock frequencies than others. Such processors work perfectly well at lower clock speeds, and are thus sold with a lower clock rating. Processors that function at higher frequencies are sold at the higher clock rating. (This is also a warning to those enterprising hardware hackers who ratchet up the clock speed of their desktop computers: the processor is probably operating out of its tolerances and will cause errors.)

The operating costs of such a complex chip foundry require that it operate continuous ly. The entire fabrication process is a living thing, in that the workers are always applying careful adjustments to the various manufacturing operations described here. These adjustments help reduce defects, and so improve the yield of the number of usable processors per wafer. The improved yields lower the overall cost of making the processors, which means more affordable and more powerful systems for the end user.


Editor's Note: The author wishes to thank Digital Semiconductor's Fab 6 for its assistance in making this article possible.


Fab Four Plus Two

photo_link (101 Kbytes)

Fab 6 's main output: Alpha processors


Not Playboy Attire

photo_link (99 Kbytes)

"Bunny suits" help control contamination. A spec of dust one quarter the thickness of your hair can ruin dozens of nascdent processors.


A Mechanized Society

photo_link (48 Kbytes)

Machines outnumber people: Robotics eve rywhere pick up and move wafers through the various processing stages.


Pattern Projection

photo_link (67 Kbytes)

A stepper projects patterns -- called masks -- onto wafers. These patterns help build the processor circuits.


An Electrifying Experience

photo_link (64 Kbytes)

An implantation unit uses an electric field to embed trace elements in silicon and change its electrical properties.


The Defective Minority

photo_link (60 Kbytes)

Rigorous testing, using machines and human inspection, finds defects in about 10 percent of the processors produced at Fab 6.


Tom Thompson is a BYTE senior technical editor at large, and has a BSEE degree from the University of Memphis, TN. He can be reached at tom_thompson@bix.com .

Up to the Features section contentsGo to next article: Building Transistors, Layer by LayerSearchSend a comment on this articleSubscribe to BYTE or BYTE on CD-ROM  
Flexible C++
Matthew Wilson
My approach to software engineering is far more pragmatic than it is theoretical--and no language better exemplifies this than C++.

more...

BYTE Digest

BYTE Digest editors every month analyze and evaluate the best articles from Information Week, EE Times, Dr. Dobb's Journal, Network Computing, Sys Admin, and dozens of other CMP publications—bringing you critical news and information about wireless communication, computer security, software development, embedded systems, and more!

Find out more

BYTE.com Store

BYTE CD-ROM
NOW, on one CD-ROM, you can instantly access more than 8 years of BYTE.
 
The Best of BYTE Volume 1: Programming Languages
The Best of BYTE
Volume 1: Programming Languages
In this issue of Best of BYTE, we bring together some of the leading programming language designers and implementors...

Copyright © 2005 CMP Media LLC, Privacy Policy, Your California Privacy rights, Terms of Service
Site comments: webmaster@byte.com
SDMG Web Sites: BYTE.com, C/C++ Users Journal, Dr. Dobb's Journal, MSDN Magazine, New Architect, SD Expo, SD Magazine, Sys Admin, The Perl Journal, UnixReview.com, Windows Developer Network