Technical Insight
Time for the compounds industry to adopt silicon concepts
The integration of silicon and non-silicon technologies is a crucial issue facing the entire semiconductor business. In order to create a successful hybrid industry, the compounds sector needs to adopt business models from its silicon counterpart while developing new technologies, as Michael Hatcher reports.
Convergence of the silicon and compound semiconductor industries promises the best of both worlds for device manufacturers - the high performance, flexibility and enhanced functionality of compounds coupled with the low-cost manufacturability and sheer scale of silicon.
GaAs may have once been regarded by many as the material of the future, but gone is the idea that it will directly replace silicon in the years to come. "Gallium arsenide will not replace silicon; the compounds industry has to learn the concepts of the silicon industry," said Alain Kaloyeros. Kaloyeros is the executive director of Albany Nanotech, a research center at the University of Albany, New York, that is investigating ways to integrate optoelectronic and microelectronic systems at the nanoscale. Kaloyeros is approaching the convergence issue from the perspective of the silicon IC manufacturer.
"Convergence is about the marriage of technological opportunities. It s about the silicon industry adopting materials and device concepts from the compounds industry and integrating them," Kaloyeros told Compound Semiconductor. "We know that compounds can bring specific benefits [to silicon], and the silicon industry is on the verge of accepting them. The biggest challenge for compounds is to get used to the silicon world."
To some extent, this is already starting to happen, with the silicon industry embracing epitaxial processes to make strained silicon and SiGe devices, but for Kaloyeros, convergence means more than this. "It means equipment vendors learning that they aren t just selling a machine, but an entire technological process, a total solution." That may sound like marketing puff, but Kaloyeros says it is critical for tool and equipment suppliers to work hand-in-hand with customers from an early stage in the development process.
Pioneering PICPerhaps the first sign that this silicon-type business model is being initiated in the compounds industry is the recent opening of Veeco s Process Integration Center (PIC). Kaloyeros told Compound Semiconductor that the PIC is a "pioneering step" that is leading the compounds industry in the right direction. "In terms of manufacturing throughput and yield, the compound semiconductor industry is 10-20 years behind silicon," he said. Kaloyeros believes that the compounds sector has to adopt these kinds of business models to create a hybrid industry. (See "Veeco s Process Integration Center: imitating the silicon business model" below.)
Another key requirement for convergence, according to Kaloyeros, is to develop the equivalent of the Semiconductor Equipment and Materials International standards that have been adopted in the silicon industry. "There are no universal standards [for equipment] in the compound semiconductor industry. The industry needs to adopt a cluster-tool approach that leads to a lower cost of ownership for the customer."
A crucial part of the process will be the drawing-up of a technology roadmap emulating the International Technology Roadmap for Semiconductors (ITRS) that is the cornerstone of the silicon industry. "In the silicon industry, the ITRS is like the Bible," said Kaloyeros. "Everybody s presentation begins with a couple of slides on the ITRS. You just don t get that in the compounds industry."
The problem for the compounds industry when it comes to roadmaps is that it is a lot more complex than silicon. Whereas silicon is primarily concerned with two key areas - logic processing and memory - compound semiconductor devices fill a much wider range of applications in optoelectronics and microelectronics.
So, rather than joining the ITRS, Kaloyeros suggests that the compounds industry develops its own roadmap, which overlaps that of the silicon industry in the appropriate places.
Serge Otkyabrsky, a colleague of Kaloyeros at the University of Albany, is looking to develop silicon-compatible compound semiconductor devices. He cites three key technology areas that are driving the two industries towards convergence. The first driver, he says, is the interconnect "bottleneck" that looms over the silicon industry.
The problem is not with increasing the speed of individual chips, to do which there are realizable technologies capable of extending Moore s Law well into the next decade. The stumbling block is to keep the individual chips communicating with each other sufficiently quickly as their processing speed becomes ever faster. The communication speed possible in metal wire connections is approaching its theoretical limit as the complexity of chips increases and their size decreases. The narrower the interconnects are, the bigger the problem of electron scattering becomes. For copper interconnects, resistivity becomes a problem when the wires reach a line width of 39 nm.
Optical solutionThe precise timing of this bottleneck is a matter of some debate, but copper interconnects are expected to continue to scale for the next three to five years before novel technology is needed. One solution would be to integrate optical interconnects at first the system level, then at the board, and ultimately at the chip level. "We need photons to solve this problem, as they don t interact with each other like electrons do," Otkyabrsky said. Being an indirect-bandgap semiconductor, silicon doesn t emit light very well, so compound devices are essential if optical interconnects are to be realized.
A key recent development in this area is the collaboration between IBM and Agilent Technologies in a $30 million funded project for the Defense Advanced Research Projects Agency. With two of the biggest companies from the silicon and optoelectronics industries working together, some exciting developments are expected.
The second driving factor, says Otkyabrsky, is the convergence of telecommunications and silicon ICs, which will demand integration of optical and wireless technology with silicon. The speed of communication is now driven largely by the speed of optical links, and to keep up, optical speeds will need to extend all the way to the chip.
According to Otkyabrsky, the third driver of convergence is the potential use of optical clocks in silicon ICs. He says that a good global clock will be crucial to maintaining good communications within current silicon IC architecture, and that an optical clock is one way to do this.
With the marriage of optoelectronics and silicon central to the convergence issue, Otkyabrsky has been concentrating on developing semiconductor light sources that are compatible with silicon.
"The problem is that current optical sources do not meet silicon requirements. For example, III-V optoelectronic devices are usually thermally stabilized, but a silicon chip gets very hot when it is in use," he explained.
Silicon chips don t heat up in a regular fashion either; instead, "hot-spots" tend to form, sometimes reaching 100 ºC. This is exacerbated by the fact that the position of the hot spots aren t precisely known. These problems will require a semiconductor laser that can withstand sudden, and potentially very large, changes in temperature.
GaAs may have once been regarded by many as the material of the future, but gone is the idea that it will directly replace silicon in the years to come. "Gallium arsenide will not replace silicon; the compounds industry has to learn the concepts of the silicon industry," said Alain Kaloyeros. Kaloyeros is the executive director of Albany Nanotech, a research center at the University of Albany, New York, that is investigating ways to integrate optoelectronic and microelectronic systems at the nanoscale. Kaloyeros is approaching the convergence issue from the perspective of the silicon IC manufacturer.
"Convergence is about the marriage of technological opportunities. It s about the silicon industry adopting materials and device concepts from the compounds industry and integrating them," Kaloyeros told Compound Semiconductor. "We know that compounds can bring specific benefits [to silicon], and the silicon industry is on the verge of accepting them. The biggest challenge for compounds is to get used to the silicon world."
To some extent, this is already starting to happen, with the silicon industry embracing epitaxial processes to make strained silicon and SiGe devices, but for Kaloyeros, convergence means more than this. "It means equipment vendors learning that they aren t just selling a machine, but an entire technological process, a total solution." That may sound like marketing puff, but Kaloyeros says it is critical for tool and equipment suppliers to work hand-in-hand with customers from an early stage in the development process.
Pioneering PICPerhaps the first sign that this silicon-type business model is being initiated in the compounds industry is the recent opening of Veeco s Process Integration Center (PIC). Kaloyeros told Compound Semiconductor that the PIC is a "pioneering step" that is leading the compounds industry in the right direction. "In terms of manufacturing throughput and yield, the compound semiconductor industry is 10-20 years behind silicon," he said. Kaloyeros believes that the compounds sector has to adopt these kinds of business models to create a hybrid industry. (See "Veeco s Process Integration Center: imitating the silicon business model" below.)
Another key requirement for convergence, according to Kaloyeros, is to develop the equivalent of the Semiconductor Equipment and Materials International standards that have been adopted in the silicon industry. "There are no universal standards [for equipment] in the compound semiconductor industry. The industry needs to adopt a cluster-tool approach that leads to a lower cost of ownership for the customer."
A crucial part of the process will be the drawing-up of a technology roadmap emulating the International Technology Roadmap for Semiconductors (ITRS) that is the cornerstone of the silicon industry. "In the silicon industry, the ITRS is like the Bible," said Kaloyeros. "Everybody s presentation begins with a couple of slides on the ITRS. You just don t get that in the compounds industry."
The problem for the compounds industry when it comes to roadmaps is that it is a lot more complex than silicon. Whereas silicon is primarily concerned with two key areas - logic processing and memory - compound semiconductor devices fill a much wider range of applications in optoelectronics and microelectronics.
So, rather than joining the ITRS, Kaloyeros suggests that the compounds industry develops its own roadmap, which overlaps that of the silicon industry in the appropriate places.
Serge Otkyabrsky, a colleague of Kaloyeros at the University of Albany, is looking to develop silicon-compatible compound semiconductor devices. He cites three key technology areas that are driving the two industries towards convergence. The first driver, he says, is the interconnect "bottleneck" that looms over the silicon industry.
The problem is not with increasing the speed of individual chips, to do which there are realizable technologies capable of extending Moore s Law well into the next decade. The stumbling block is to keep the individual chips communicating with each other sufficiently quickly as their processing speed becomes ever faster. The communication speed possible in metal wire connections is approaching its theoretical limit as the complexity of chips increases and their size decreases. The narrower the interconnects are, the bigger the problem of electron scattering becomes. For copper interconnects, resistivity becomes a problem when the wires reach a line width of 39 nm.
Optical solutionThe precise timing of this bottleneck is a matter of some debate, but copper interconnects are expected to continue to scale for the next three to five years before novel technology is needed. One solution would be to integrate optical interconnects at first the system level, then at the board, and ultimately at the chip level. "We need photons to solve this problem, as they don t interact with each other like electrons do," Otkyabrsky said. Being an indirect-bandgap semiconductor, silicon doesn t emit light very well, so compound devices are essential if optical interconnects are to be realized.
A key recent development in this area is the collaboration between IBM and Agilent Technologies in a $30 million funded project for the Defense Advanced Research Projects Agency. With two of the biggest companies from the silicon and optoelectronics industries working together, some exciting developments are expected.
The second driving factor, says Otkyabrsky, is the convergence of telecommunications and silicon ICs, which will demand integration of optical and wireless technology with silicon. The speed of communication is now driven largely by the speed of optical links, and to keep up, optical speeds will need to extend all the way to the chip.
According to Otkyabrsky, the third driver of convergence is the potential use of optical clocks in silicon ICs. He says that a good global clock will be crucial to maintaining good communications within current silicon IC architecture, and that an optical clock is one way to do this.
With the marriage of optoelectronics and silicon central to the convergence issue, Otkyabrsky has been concentrating on developing semiconductor light sources that are compatible with silicon.
"The problem is that current optical sources do not meet silicon requirements. For example, III-V optoelectronic devices are usually thermally stabilized, but a silicon chip gets very hot when it is in use," he explained.
Silicon chips don t heat up in a regular fashion either; instead, "hot-spots" tend to form, sometimes reaching 100 ºC. This is exacerbated by the fact that the position of the hot spots aren t precisely known. These problems will require a semiconductor laser that can withstand sudden, and potentially very large, changes in temperature.
