Chipmakers make smooth shift to copper

Copper, which conducts electricity better than aluminum, gives designers an avenue to break through looming physical barriers that could prevent further boosts in chip performance. The first copper Pentium 4's will come out in the fourth quarter of this year at 2.2GHz, for instance, and hit 3.5GHz next year.

Working with copper poses several challenges, however. "Sputtering," a process for applying metal to silicon, doesn't work with copper, for example. Neither do traditional techniques for etching circuits. And errant, minute traces of copper rubbed on a wafer can destroy a batch of chips.

Analysts predicted that production hiccups could result in annoyances for medium-sized producers or in financial disasters for larger ones.

"It was pretty scary, frankly, at the beginning," said Mark Bohr, an Intel fellow and director of architecture and integration.

Nonetheless, the conversion has been unnaturally quiet. IBM, which released the first copper chips in 1998, is almost all copper now, and Advanced Micro Devices started churning out copper Athlons last year without incident. Intel, Taiwan Semiconductor Manufacturing Co., Via and Sun Microsystems, among others, have all launched their first copper wares in recent months, with volume production to follow soon.

Speed, or lack of it, was a huge factor in the change. IBM started performing copper experiments in the mid-1980s. IBM and Intel also coordinated efforts with equipment makers such as Novellus to ease the transition to mass manufacturing.

"Because most of the companies implemented it very slowly, the equipment industry kept up and had a solution in place," said Jim Ryan, an IBM distinguished engineer and manager of the company's interconnect technologies. "It wasn't available in 1998, when we went to full-scale manufacturing, but it is now."

Noted Dean McCarron, principal analyst at Mercury Research, "The transition has gone much smoother than anticipated."

Just as important, companies rethought their manufacturing processes to completely isolate the metal and get around the contamination issue.

"There are whole bays in the (chip plant) where signs say 'No Copper Allowed Past This Point,'" said Nathan Brookwood, an analyst with Insight 64. "It completely screws up the semiconductor operation of silicon. Aluminum is relatively inert."

Why copper?
The switch to copper became a necessary consequence of chip engineering. Under Moore's Law, the number of transistors on a given semiconductor doubles roughly every 24 months. The doubling occurs largely because the transistors are continually being shrunk. Smaller transistors improve performance by allowing designers to cram more features onto a single chip. In addition, signals travel faster because electrons don't have as much real estate to cross.

For years, designers primarily could improve performance by reducing the distance between transistors. The shrinking size of wires, however, began to create problems. Smaller wires carry less current and are more resistant to electricity.

In the chip generation before going to copper, Intel partly ameliorated the problem by changing the "aspect ratio," or horizontal-to-vertical shape, of its aluminum interconnects. Still, a change in materials was inevitable. Copper has greater current density than aluminum, and its resistance is 30 percent to 40 percent lower.

"The amount of the total delay that was due to the interconnect system was getting larger," IBM's Ryan said. "Copper really was the only real choice. Silver and gold were more expensive than you wanted to deal with."

Copper is also less prone to electromigration, added Intel's Bohr. A dense, constant flow of electrons across an aluminum wire can dislocate the metallic atoms and create a void, which results in chip failure.

Unfortunately, aluminum and copper don't behave the same way. Aluminum can be applied to chips through a chemical vapor. Not so with copper.

"You could sputter it, but that kind of spewed the copper all over the place," said Ryan. Eventually, IBM centered in on electroplating, in which metal is adhered to a surface through an electronic charge.

Etching, the process of laying circuits on wafers, also had to be altered. With aluminum, metal gets sprayed across the entire surface of a wafer. Metal is then etched away until only the wires that form the circuit pattern remain.

Dig that process
For copper chips, designers went to the "dual damocene" process. Originally developed for tungsten chips in the early 1990s, the dual damocene process requires that manufacturers dig the circuit pattern in trenches onto the wafer. Copper is then electroplated across the entire surface. After etching, the metal left in the trenches forms the circuits.

Another problem: contamination. "If any copper gets on a wafer, it will quickly diffuse through it," Bohr said. "Aluminum will not diffuse rapidly. You could get minute traces on the back of a wafer" and still produce chips with it, he added.

Designers came up with adhesive, insulating layers, usually made out of tantalum, to separate the silicon oxide and the wires but keep them stuck together. Factory floors also were rearranged.

"With aluminum, all the process steps could be intermingled," Bohr said. "We solved this by putting different tools in different parts of the clean room...There were no big 'gotchas,' but for about a year or so there were lots of little problems."

The first chip to use copper, a 400MHz Power PC from IBM, came out in September 1998. Except for a few occasional glitches, the conversion has largely succeeded across the industry.

"It's actually less expensive to work with copper. You can eliminate a few steps," Insight 64's Brookwood said.

Added Kevin Krewell, an analyst at Microprocessor Report: "When you change materials, it can be scary, but copper is not esoteric. If you handle it right, you do OK. Guys like IBM knew their stuff, and they got it right."

The next generation of problems, though, is already looming. Further shrinkage will require more and better insulating techniques. Advanced semiconductor manufacturing techniques will also have to be applied to chip packaging, which contains the wires, which connect a semiconductor to other components.

Ryan, among others, is optimistic. "The key ingredient is the determination to make a change," he said.