Casual observers point to Dell Inc. to make the argument that computer design is effectively dead. But a close look at the current practice and the history of engineering at Dell yields a somewhat different conclusion.
Dell has been, and continues to be, a champion of innovative design, though the company practices it in a highly minimalist manner and believes that the field for electronic innovation in PCs has drastically narrowed. More important, Dell has helped define a new role for the OEM system maker in a mature, disaggregated industry.
That role is all about owning deep insights into the end customer, and using that information to influence its suppliers' silicon designs. Now that competitors such as archrival Hewlett-Packard Co. are practicing similar methods, Dell is looking to up the ante by injecting a new level of formalism into its processes.
Dell sets the pace for how little a top computer maker can spend on design. The company generated record net revenue of $49.2 billion in its last fiscal year (ending Jan. 28, 2005). That same year, it reported record-low expenses for research, development and engineering combined of just 0.9 percent of net revenue.
While the company claims it employs about 4,000 engineers of all types among its 63,700 staff members, EE Times estimates fewer than 250 of them are EEs working on homegrown designs. Based on interviews with top engineers, well less than half of the company's systems probably fewer than 100 or so board-level products a year are designed in-house.
Rather than gloat over its Zen-like achievement, Dell's top technology execs are almost defensively quick to champion the company's history of innovation. In an interview, Jeff Clarke, senior vice president of Dell's products group, rattled off a short list.
In 1985, Dell rolled out the industry's fastest 286 PC, thanks to work debugging a third-party chip set. In 1988, it scored the fastest 386 by incorporating SRAM memory to eliminate wait states. Dell shipped a 486 that linked graphics directly to the CPU before the industry defined a processor local bus. It was the first to incorporate lithium-ion batteries and, later, a fast-charge scheme in a notebook. More recently, Dell helped define a new RAID format to eliminate incompatibilities dogging storage chips.
Clarke is quick to admit a PC maker's innovations today are more in areas such as software and systems interfaces than in core silicon. "You don't need more guys designing silicon and disk drives, you need guys integrating this into systems," he said.
An EE and a Dell employee since 1987, Clarke has driven much of Dell's technology strategy for years, according to insiders. "He believed we could do designs that were as cheap as those from Taiwan, but better, and still have some customization," said Jay Bell, a retired Dell fellow.
Chief executive officer Michael Dell hired Bell as a hotshot design consultant in late 1985, challenging him to deliver the fastest 286 system in time for the Comdex trade show the following spring. Within three weeks, due to an obscure bug in a logic chip that only he found, Bell had a prototype running at 12 MHz at a time when other computers topped out at 8 to 10 MHz.
For just $90 more in components, Dell was able to sell a PC with a price tag several hundred dollars above that of its next-fastest system. "It was a big moneymaker" and impressed customers such as Exxon and the Jet Propulsion Lab, Bell recalled.
Unlike anywhere else
Although his degree was in Germanic languages, not engineering, Bell became one of about seven designers at the company. "Engineering at Dell was not like engineering anywhere else. People sometimes weren't really engineers," Bell said.
That started to change in the summer of 1988, when Michael Dell recruited IBM Corp. fellow and his Austin, Texas, neighbor Glenn Henry to build a proper engineering team. IBM had just killed Henry's initiative to build an X86 clone CPU but not his fire to get involved in the emerging PC industry.
"An IBM fellow is a nice job with a lot of perks and security, [but] I decided I was going to be a PC guy," Henry said. "And Dell was small. That appealed to me."
By the time Henry left the company in late 1993, Dell had grown from 700 people and $200 million in revenue to 2,000 employees and $2 billion in sales. The engineering group expanded from a few people to a few hundred. Along the way, the company made a strategic decision not to develop any of its own chips after an ambitious project that Henry had championed failed.
Project Olympus was a memory controller that let users upgrade a single system from 386 to 486 and i960 CPUs, as well as ISA, EISA and MCA add-on boards.
"We were trying to come up with a modular family that could use all the CPUs and buses with a common set of components," said Henry. "Practically, when you looked at the costs and the infrastructure of what was really out there, it made no sense."
Parallel chip and board design efforts stretched out two years, during which Dell had few new designs to show. In late 1989, "Glenn finally had to kill his own program," said Bell.
After a stint at PC chip set company Zymos, Bell returned to the company and helped drive the decision to cancel Olympus and instead use core logic from relative newcomer VLSI Technology. It was something of a turning point for both Dell and Henry.
"We decided to stop developing some of the boards and chips and concentrate on systems integration," Henry said. "I wanted to do and manage hard-core technology work, but as the company grew it was clear there would be less and less opportunity to do [that]." Henry is now CEO of X86 CPU designer Centaur Technology Inc. (Austin).
"It was during this time that Intel got into the motherboard business and Taiwan became very aggressive in board design," Henry said. "The boards became more and more integrated, and there was less and less opportunity to add value. We could get anything we wanted from other companies.
"Dell's primary value when I left was to specify what the systems should be, choose partners to develop components for them and do the integration and test. It became more like technical project management," Henry added.
As Henry grew disillusioned, Bell was honing the art of influencing chip makers, increasingly by using information Dell gleaned from its customers about the features and prices they wanted to see. His big success was helping National Semiconductor Corp. see the wisdom in rolling up a host of discrete floppy-disk, hard-disk, serial and parallel controllers into a single chip that became known as the Super I/O, a significant moneymaker for National.
"That was my effort for three or four years," said Bell. "We even tried to get the Ethernet controller in there, but it turned out to be the straw that broke the camel's back."
In 1994, Bell became the DRAM guy at Dell, traveling to Japan and South Korea three times a year. Thanks to Dell's growth and its direct-sales model, the company established a track record of placing big and reliable orders as competitors over- or underbought parts during cyclical shifts in memory demand. "By '98, we had a lot of credibility with memory vendors," Bell said.
Today, chief technology officer Kevin Kettler runs the 90-person architecture group at Dell, responsible for influencing all the computer maker's major chip and software vendors, including Intel and Microsoft.
"We try to have a single point of accountability for key technologies such as memory, graphics and so on. These people have authority to make decisions across the product line," Kettler said. "Vendors say it's great getting a single voice for Dell."
Dell decides on the major chips for its products in a series of meetings that include members of Kettler's architecture group as well as product-engineering, marketing and procurement managers. The teams share the pros and cons of the components and their vendors to reach a consensus, Kettler said.
One of the ways Dell seeks leverage with it suppliers is by sharing its extensive data on end-user demands. "We spend an enormous amount of time with customers" in various summits, advisory councils and one-on-one meetings that the company has made a regular part of its business for the last 10 years, said Kettler.
Dell conducts quarterly customer advisory councils by product sector, in which as many as 25 people are "locked in a room for two and a half days to tell us everything they do and do not like about our products and our road map," said Forrest Norrod, a vice president for enterprise engineering at Dell.
From these meetings, Dell execs draw up a list of "behavioral specs" that define customer requirements and are used as part of final system tests.
"It's quite a bit more formal" than a simple list of features, said Norrod, and might include the style or implementation of a feature, such as the specific brightness of a front-panel LED.
Dell is starting to inject more-formal methods of quantitative data collection into its user research, such as measuring when and how users upgrade server components, down to the amount of force they use to insert a plug-and-play drive into a rack-mounted system, said Norrod.
"We have an almost fanatical view of engineering decisions being based on defined processes and metrics," said Ashley Gorakhpurwalla, who manages a group of about 60 Dell engineers who design the server boards developed in-house. "Every decision becomes technical if you have the data you need."
Dell designs about half its servers in-house and outsources design of the rest. In total, the server group builds about 60 boards a year, some for systems that are otherwise designed outside the company. "We do everything from simulation to physical design and layout and ECAD," said Gorakhpurwalla.
From its chip vendors, Dell gets buffer models showing electrical-interface and signaling parameters. Engineers typically use simulation, schematic capture and layout tools from Cadence Design Systems Inc., although some use schematic tools from Mentor Graphics Corp. Dell has developed its own software to handle many test functions and to automatically generate a bill of materials.
Norrod said Dell has a "fairly invasive engagement" with outside design partners who develop the rest of its products. "We specify the architecture to a pretty low level. We usually get down to which bridge chips to use, what kind of interrupt routing and what I/Os," he said.
A small sampler of Taiwan-based board makers paint a different picture, though none were willing to be interviewed for the record. Dell pretty much adopts the Taiwan-designed boards and their bills of materials, and spends most of its engineering effort verifying quality and compatibility, the Taiwanese sources said.
Whatever the reality, computer design is still getting done even if only a small fraction of it is now at the big OEMs. "There are really two computer industries," said Henry. "There are the Dells and HPs, and then there are the board and chip makers a million infrastructure companies where there is real design activity."