By Dave Lammers
Keynote speakers Terry Higashi of Tokyo Electron Ltd. and Tom Caulfield of GlobalFoundries took the stage at the Yerba Buena Theater Tuesday morning to predict major changes in the goals and operations of the semiconductor industry.
In many ways, 2017 has been marked by intense interest in the capabilities of neural networks and other forms of artificial intelligence (AI). Higashi, now a corporate director at TEL, predicted that AI and virtual reality are among the applications that will propel demand for semiconductors “almost without limit.” Neuromorphic processors, the veteran TEL executive said, “are one of the promising devices to enhance human creativity. They will be improved step by step, just as logic and memory devices were improved.”
Looking toward a future in which AI and human skills combine to resolve problems, Higashi predicted that today’s Von Neumann-based architectures and neuromorphic device will complement each other. “Artificial intelligence solutions will be proposed, and the challenges and problems will be solved by scientists and engineers. The combination of Von Neumann and neuromorphic computing gets us closer to true intelligence,” he said.
AI also will play a role in enhancing the immersive experiences promised by virtual reality, experiences which visionaries have predicted but which thus far mankind “has never fully experienced.”
Higashi said that by combining VR and AI, “we can attain a suspension of disbelief, and simply enjoy the experience. If we can provide the technologies, consumers will experience excitement and a form of happiness.”
Caulfield, the general manager of the Malta fab near Albany, agreed with Higashi’s assessment that that the semiconductor industry is seeing “new buds” that will bloom into large semiconductor markets.
However, Caulfield said that to achieve anything like the rate of technological progress seen over the first half century of the semiconductor industry, companies and customers will have to take collaboration to new levels. And he offered the collaboration between GlobalFoundries and AMD as an example.
“Collaboration, potentially, is the biggest thing we need to do. We need strategic partnerships, and not only among semiconductor manufacturers but also with equipment suppliers.”
At its Malta fab, GlobalFoundries builds all of AMD’s leading-edge discrete graphics engines and CPUs. “The AMD and GlobalFoundries engineering teams are so embedded with each other, one can hardly tell” which company an engineer works for, he said.
Noting the resurgence of AMD, Caulfield said “we are all proud to be part of that partnership.” And he pointed to another collaboration, between Samsung and GlobalFoundries, which allows customers to take the same 14nm design and choose whether to manufacture it at Samsung’s Austin fab or at Malta. “Customers can run photomasks in Austin or in Malta, New York and have the product look the same,” he said.
Government role
In such a collaboration-rich business environment, governments also have a role to play, Caulfield said.
“Public-private investments must imply a return to governments as well as to companies. Otherwise, they send the wrong message.” By investing several billion dollars in the Malta fab, GlobalFoundries and the state of New York put to work the well-educated young people who otherwise would have left the state in search of technology jobs. When Malta began operations, only 20 percent of the staff were educated in New York. Now, fully half of the workforce has benefited from a New York education.
“We were exporting talent. Now, the workforce has great opportunity within the state,” he said.
Both Higashi and Caulfield said major challenges face the industry. Higashi noted that innovation will be required to keep flash memory costs under control. “As data is captured by sensors and is transferred via the appropriate networks and stored in data centers, demand for NAND will be high. We must make huge efforts to reduce the overall cost, as the semiconductor industry is expected to provide enough volumes to support the Internet of Things.”
Caulfield said the performance of logic transistors has struggled to keep pace, even as density increases have continued. When the industry moved from 28nm to 14nm technologies, performance increased by fully 50 percent. But from 14nm to 10nm, speeds improved by about 18 percent, making shrinks primarily a cost improvement.
With the industry now focused on brining 7nm logic to the market, the question arises whether 5nm CMOS will provide enough performance to justify that node. While the jury on technology scaling is still out, Caulfield said the industry may have to move to gate all around (GAA) structures, or to non-silicon channel materials, in order to gain the kinds of performance improvements that customers expect from a new node.
Higashi said systems must get faster. “Real-time processing is crucial in the cyber world. And with robotic hands, there should be no delays in physical operations.”
“Memory, logic, and sensing make it possible for AI systems to solve problems much faster than a team of geniuses. We are now in a new era, one of super integration. In addition to improved specialty devices – based on logic, memory, and sensors – we must take these separate devices and put them together into fully integrated systems. It is time to make a pizza, with some of the best ingredients,” he said.
