Because of the need for atomic-level precision and the high risk of contamination, the 200-mm silicon wafers onto which semiconductors are built are handled almost exclusively by machines. To move from machine to machine, the wafers are loaded into plastic cases and shuttled along a mile-long enclosed monorail that snakes throughout the plant.
About 300 of the 750 employees at Motorola's plant in Chandler are operators who load cases of silicon wafers into machines for processing. These operators are trained to run several machines in the same area to reduce the time spent walking. In one instance 2.5 hours of daily process time was eliminated by taking this "staffing by geography" approach.
Manufacturing as it is practiced at Motorola's semiconductor fabrication plant in Chandler, Arizona, more closely resembles a nuclear production facility in a James Bond movie than what one typically sees in an auto factory. Operators and technicians outfitted in clean room "bunny suits" load and monitor gleaming stainless steel machines that process materials with a precision measured in atoms. The microprocessors that are the chief product of this $1.5 billion facility could fit on a fingernail, but are the key to practically every new innovation planned in the automotive industry from telematics to X-by-wire systems.
How To Make A Chip
The Chandler plant receives 200-mm discs of silicon called "wafers," which resemble oversized black compact discs. These wafers provide the base onto which between 150 and 600 chips are built. The manufacturing process that makes the chips is a cross between photo processing and high-energy physics with a trip to the car wash thrown in.
The wafers are treated with a photoresistive material and then exposed using satellite-grade optics to create intricate patterns with elements as fine as 0.18 micron. This pattern provides the roadmap along which current will run in the completed chip, and its intricacy determines the number of transistors that can be fit on one chip, which in turn roughly determines computing power. (Currently deep ultra-violet light is used for exposure, but the semiconductor industry is researching using X-rays which could further increase the fineness of the patterning.) The exposed areas are etched with chemicals to create holes into which accelerated ions are implanted at top speeds of 7.9 million mph. These ions, which include aluminum and silicon dioxide, are chosen for their relative conductivity and form the actual pathways along which current flows. Once these pathways are in place, the photoresistive material is stripped away and the process begins all over again, eventually resulting in multiple layers of different but interconnected patterns.
The entire manufacturing process covers 120 operations with over 350 processing steps. In addition to the steps mentioned above, each wafer is meticulously cleaned and polished several times during its journey through the plant using chemicals that aid in etching and an abrasive slurry containing 0.1- to 0.2-micron silica grit. Since elements like sodium in the water used for the cleaning process can cling to the microstructure of the chip and create unwanted conductive pathways, the plant uses ultra-pure water that is generated on-site using reverse osmosis. Contaminants in the water are reduced to a parts-per-trillion level before it is deemed safe to be used in chip cleaning operations.
Given that Motorola is dealing with literally microscopic dimension tolerances, the opportunity for creating defects is immense. When a dust particle that measures only 0.1 micron can cause a chip to fail, there is no opportunity to fix problems after the fact. So each production machine is both continuously self-monitoring and self-adjusting. And operators are trained to detect the early warning signs of possible problems and communicate them up the management chain immediately. There is also such a tremendous emphasis placed on ensuring strict standardization of procedures that it might surprise even a Toyota engineer. Ken Williams, operations manager at the plant, puts it this way, "The essence of this industry is to remove all variation."
One way that Motorola attempts to do that is by requiring that every engineer be trained to near black-belt levels in statistical process control (SPC). Even maintenance and process technicians and manufacturing supervisors must study SPC, though at somewhat less rigorous levels. Also, 11 cross-functional teams have been set up to focus on improving quality and increasing the yield rate for good chips using the "8D" problem-solving methodology familiar to the automotive industry.
Plant personnel get a jump on fine-tuning their processes for the next generation of chips because they share equipment with an on-site advanced research lab. This partnering came about mainly out of economic necessity since the machines needed for both production and testing are so costly (some are $8-million a piece) that having a separate and fully equipped test plant is not economically feasible. But beyond saving investment dollars, this arrangement allows chip developers to interact with production personnel on a daily basis, enhancing communication and reducing the potential of manufacturing problems when the new chips begin mass production.
The Most Stringent Standard
Unlike many chip makers, Motorola does not grade chips based on application, where lower performing units go into lower end products. At Chandler, every chip must meet the highest test standards regardless of whether it is going into a cell phone, a pager or an automobile. Williams says, "We have chosen to hold ourselves to the most stringent standard – automotive." This means withstanding a temperature range from -40°C to +125°C and dealing with the vibration and harshness of the engine compartment environment. Automotive applications require tests that are designed to weed out weak units and those rigorous tests drive the need for extensive SPC measures. Over the years, these efforts have kept the defect rate for chips essentially the same, which on the surface doesn't sound like anything to boast about. But considering that there were only 4,000 transistors on the first automotive chips launched in 1978 and that there are more than 7 million in the current generation, keeping defect levels the same has meant actual quality increases approaching 2,000 times better. This has allowed Motorola to remain competitive in an automotive arena that is constantly driving costs down. "We aren't so much cutting prices as increasing performance per dollar by orders of magnitude," says Roger Tyldesley, device engineering manager at the plant.
Since the production process at Chandler cycles wafers through the same processes several times and is not linear like a vehicle assembly line, there is greater opportunity to make changes on the fly without disrupting production. The plant's computer integrated manufacturing system re-prioritizes wafer lots every 11 minutes based on the latest order and scheduling information. It analyses factors like bottleneck toolsets and real-time yield rates and adjusts wafer lot order accordingly. It also monitors scrap rates and re-distributes wafers between orders to fill the highest priority orders first. A key component of the system is a manufacturing execution system solution called PROMIS provided by PRI Automation, Inc. (Billerica, MA). But the overall system is an integration of many solutions that was customized at the plant and has become a touchstone for other Motorola facilities.
The demands placed on the Chandler plant will increase in the future as the level of sophisticated electronics in vehicles increases. According to Peter Schulmeyer, strategic marketing director for Motorola's Transportation and Standard Products group, "Much stricter emissions standards and expectations for fuel consumption and engine performance are driving up the computational performance requirements of engine management dramatically." And telematic systems require even higher performance levels than engine management. Additionally, X-by-wire controls and safety systems like video-equipped air bags that adjust their deployment based on the evaluation of a video image will require tremendous amounts of memory.
Because automakers require such a long and detailed qualification cycle, the chips that go into cars lag one to two generations behind those inside the latest cell phone. And modifications needed to meet operating temperature and low power consumption requirements reduce automotive chip performance by a factor of five below other products. Still, with chips on the drawing board that will increase the number of transistors from 7 million today to 40 million in a couple of years, Motorola should make up for much of the performance difference while keeping automotive chips the gold standard for reliability. And reliability will remain paramount. After all, a malfunctioning engine management chip is potentially far more problematic than the inconvenience of having to re-boot your laptop after a lock-up. Schulmeyer sums it up this way, "Consumer expectations are much higher for automobiles. They must never fail."