For half a century, Moore's Law has charted semiconductor development with incredible accuracy. In 1965, Gordon Moore quantified the scaling principle for semiconductors by predicting that the number of transistors that can be etched on a computer chip would double every one year. He revised it to every two years in 1975. His initial belief was that this would hold for a decade. But five decades later, Moore's Law has held up surprisingly well. However, there is now a belief among many semiconductor engineers and scientists that we will hit a miniaturization wall unless there are major breakthroughs in advanced materials innovation.
Looking back to the 1960s, one transistor cost approximately $8 in today's dollars and there were less than 1,000 transistors on a chip.Companies like Fairchild and Intel (Gordon Moore's company) started making semiconductors in Silicon Valley. They would form the basis of a computer industry that we know today.
To put things in perspective, the iPhone 6 has many more times the computing power and memory than all of NASA back in 1969, when it put a man on the moon. And more than the largest IBM mainframe computer of the day, the Multiple IBM System/360 Model 75 mainframe costing up to $3.5 million.
The bi-annually doubling of transistor density was largely made possible because of a shrinking of the transistor size and the scalability of semiconductor fabrication facilities. This traditionally worked well to increase performance and reduce cost.
The size of semiconductor nodes is commonly used to illustrate the size of transistors. Semiconductor fabs are currently producing at the 22 nm (nanometer) and 14 nm node size. This is incredibly small. By comparison the average width of a human hair is 80,000 nm.In his keynote speech at the Semi Strategic Materials Conference in Silicon Valley on September 22, Gary Patton, CTO of GlobalFoundries, discussed the history and future of semiconductor development. He stated that prior to the 90 nm node size, performance improvement was principally achieved from scaling. However, as we have moved down to 22 nm and 14 nm nodes, improvements have come from innovation in advanced materials.
Historical examples of materials innovation that have increased semiconductor performance are Cu interconnects in 1998, SiCOH based low-k dielectrics in 2002, NiPtSi-based contact materials in 2005, embedded SiGe Epi in 2006, stress liner materials in 2007, immersion lithography in 2010, high-k metal gates in 2011, and FinFET transistors in 2014.
But something unexpected occurred at 28nm. The cost per gate flattened out after dropping for years. This was because the complexity of device fabrication skyrocketed. Gary Patton stated in his speech "The era in which shrinking features automatically ensured cheaper transistors is over!" Most of the near term solutions to keep Moore's Law on track are incredibly expensive.
For example, the major breakthrough at 20nm was double patterning to "fine tune" the nodes.Basically, single lithography patterning wasn't good enough at this node size and it had to be done twice. To get below 14nm, some foundries are contemplating quadruple patterning.Can you imagine the cost? Others are contemplating a new technology called extreme ultraviolet (EUV) lithography. This technology uses extreme ultraviolet wavelength light, expected to be 13.5nm. It is created from a laser-induced plasma sourced from a lightning bolt's worth of electrical current. This is an extremely fine light source that can do lithography patterning down to sub 10 nm nodes. ASML makes these machines. The hitch is that each machine costs approximately $75 million.In April 2015, Intel ordered 15 machines for over $1 billion.
The semiconductor industry may be the most complicated industry from a pure science perspective. It involves billion dollar fabrication facilities and machines that manipulate material at the atomic level. Fabs are now using atomic layer deposition, controlled atomic layer etch, atom level planarization, atomically pure materials, and atom-level characterization equipment. That is a mouthful but it is clear that semiconductors involve deep science and expensive technology.
Looking ahead, for Moore's Law to hold, it is generally felt that the semiconductor industry will need to move to 10 nm, 7 nm, and below. I mentioned some of the near term technologies above. But further out, we have seen early-stage companies working on things like silicon nanowires, carbon nanotubes, photonics, 3D multi-chip stacking, and quantum computing. IBM recently announced a carbon nanotube transistor that scales down to 1.8 nanometer nodes. (EE Times "IBM Nanotubes May Redefine Future of Moore's Law" October 1, 2015) These may be seen as "out-of-the-box" solutions, but some very smart people are working on them and it is possible that they will be in your future handheld device.
At Pangaea Ventures we believe Moore's Law has legs. It will continue for at least another decade largely due to advanced materials innovation.