Cu vs Al

Cu vs AlIn the copper versus aluminium debate, Intel is taking an idiosyncratic stance. Will other industry bosses seek to maximise their investment in aluminium by following Intel’s example? asks David Manners
Intel is resisting the Copper Rush and reckons it can make aluminium last until 2002. Most of the rest of the industry leaders want to move to copper this year or next.
Last year IBM began shipping the first commercially available chips with copper interconnects – the 400MHz PowerPC 750 microprocessor. IBM says copper delivers 20 to 30 per cent more speed and a ten to 12 per cent reduction in power consumption compared to aluminium.  
“All our Asic designs will be on copper by the end of 2000,” says David Cremese, field engineering manager for Europe the Middle East and Africa at IBM Microelectronics. IBM has 23 copper Asic designs currently going through its fab with densities up to five million gates and says it has finished chips which are fully working. Quarter micron is IBM’s last aluminium process generation and six other companies either have, or expect to have, copper interconnect processes up and running by the end of this year.
But Intel believes it can get just as good results at the same level of process technology using aluminium. “Copper technology is not yet at the point where it is cost-effective,” insists Pierre Mirjolet, Intel’s architecture marketing manager, “the equipment is immature and the yield is worse than aluminium.”
Fabricating interconnect represents 50 per cent of the cost of making a chip, explains Mirjolet, and to make copper interconnects, you need new equipment to make the layers, to etch them and to planarise them. However companies wishing for an economic transition from 0.25 micron to 0.18 micron will want to re-use as much equipment as they can from the 0.25 micron generation. Hence Intel’s wish to stick to aluminium.
Both IBM and Intel refer to their latest processes as ‘0.18 micron’. However, by using phase shift mask techniques, Intel is achieving effective gate lengths of 0.13 micron and IBM is achieving 0.12 micron effective gate lengths.
IBM’s copper process has six layers of metal, an operating voltage of 1.8V and can put 200 million transistors on a chip. Intel’s aluminium process has six layers of metal, 3nm oxide thickness, shallow trench isolation (rather than the older LOCOS), low resistance silicide layer, 1.1-1.5V operation and can put more than 100 million transistors on a 207mm2 chip.
“We’ve got one production line at Burlington Vermont running copper in quantities of tens of thousands of wafers a month and a pilot line at East Fishkill where the process was developed,” says IBM’s Cremese. Intel’s process is being ramped into full production volumes at the Oregon fab where it was developed and, within twelve months, will be installed at two more production fabs.
IBM’s process is delivering five million gate Asics running at 500MHz and an ISSCC paper this year describes an Silicon-on-Insulator/copper version of the PowerPC750 which runs at 550MHz. Intel’s process has delivered a 900MHz 16Mbit SRAM. “Before the year’s end it is very likely it will demonstrate gigahertz microprocessors,” says Mirjolet. Intel expects it to be their workhorse technology until 2002.
A main reason why Intel remains wedded to aluminium despite the 40 per cent resistivity advantage of copper is because, according to Mirjolet, aluminium interconnects can be made thicker (higher off the substrate) than copper interconnects. “The higher you make the interconnect the less the resistance,” says Mirjolet. That outweighs much of the resistivity advantage of copper.
Nonetheless, many of the top semiconductor companies are following IBM, not Intel. Motorola was not far behind IBM in producing a copper-interconnect IC, which was its version of the PowerPC. Besides IBM and Motorola five companies expect to have copper processes in place this year: NEC, VLSI Technology (both on 0.15 micron) TSMC, UMC and Samsung (on 0.18micron). Next year AMD, Philips/ST in Grenoble, TI and Siemens are expected to have 0.18 micron copper processes. Intel concedes it will eventually go to copper – when it moves from 248nm wavelength steppers to 193nm steppers and low-K dielectrics – expected in 2002.
Copper is only useful, at the moment, for advanced logic ICs with their increasing complexity and number of metal layers. Many companies will initially use copper interconnects only for the upper metal layers, which are normally used for longer, critical-path routes. For metal layers one and two, chip designers are seeing little performance difference between aluminum and copper.
According to IC Insights of Arizona, the advantages of copper interconnect are: lower resistivity and capacitance; higher electromigration resistance; better step coverage; it is deposited at lower temperatures; it has better thermal conductivity. Disadvantages are: it is difficult to etch; it is vulnerable to scratching and corrosion; its poisonous to silicon so requiring barrier materials; it is prone to erosion from oxidisation at low temperatures.
According to Sematech of Texas, the international consortium which is co-ordinating production equipment development for new processes, a chip with 0.10 micron design rules would require 14 levels of metal using aluminium. Switching from aluminum to copper would reduce the number of layers by two; switching to both copper and low-K dielectrics would reduce layers by five or six.
While top manufacturers are expected to switch to both copper interconnects and low-K dielectrics, there are differences in implementation: IBM and NEC are moving first to copper interconnect and later to low-K dielectrics; Intel, LSI Logic, Lucent, and Toshiba will go first to low-K dielectrics and later to copper.
Intel’s example is powerful. If it can get away with aluminium for another generation, a lot of chip industry bosses will be thinking: ‘Why can’t we?’

Leave a Reply

Your email address will not be published. Required fields are marked *