The semiconductor industry is acting quite young for its age.
And that’s just not me saying it; over and over again during SEMICON West 2017 I heard some variation of the phrase “We’re just getting started.”
What was it that accounted for the upbeat and buoyant mood at SEMICON West 2017 this month, despite Moscone South Hall being a gaping hole clear down to the basement as a result of South Hall’s demolition and reconstruction?
Maybe it was that, not having the South Hall available, SEMI squeezed and rearranged the show floor from the configurations of prior years and, with everyone packed a little tighter, and with the crowd a little disoriented as to where familiar exhibitors and show services were to be found, it seems like the event’s complacency was shaken up a bit, in a good way.
That, and it being the record year for semiconductor capital equipment ($49B estimate for CY2017, topping the last high-water mark, in CY2000), could be the other factor accounting for the show mood.
My observations, after three days at the show (and the previous day at the Heterogeneous Integration Roadmap Committee all-day working meeting on Monday the 10th), are that a few major themes dominated: Heterogeneous Integration / Advanced Packaging / Fan-Out Packaging; Artificial Intelligence; Autonomous Vehicles; China’s Rise and its Future Importance to the Semiconductor Industry; Smart Manufacturing; and Sensors and the IoT. (Shades of the ASMC 2017 panel discussion!)
(Kudos btw to @SEMIexpos for the excellent and useful #SEMICONWest 2017 smartphone app – good stuff! Daily schedules, #SocialMedia, speaker bios, etc., all in one convenient place.)
As a nod to the kind of importance advanced packaging is now being accorded, the SEMI awards at SEMICON West 2017 went to Micron Technology and the HMC Consortium for commercializing the Hybrid Memory Cube 3-D stacked DRAM product, and also went to AMD for their Fiji SiP product, which utilizes a 2.5D silicon interposer and stacked DRAM. Both the Micron and the AMD products rely crucially on Through-Silicon-Via (TSV) process technology, long in development, and now widely deployed.
For Dr. Babak Sabi, Corporate Vice President, Director of Assembly and Test Development Technology, Intel, heterogeneous integration is not an “If” – rather, it’s a “Must.”
Dr. Sabi said that “Fan-Out packaging will eventually come around” to the point where fan-out is capable of creating the smaller-geometry lines and spaces that can currently be done on silicon interposers (think 0.5µm lines and spaces), and that fan-out panel level packaging will come around too, perhaps three years behind wafer level, maybe as a result of implementing process steps like CMP-based planarization to improve panel flatness and, thereby, photolithography resolution.
And all Dr. Sabi wants is processing tools capable of front-end performance, at back-end equipment prices … We’re just getting started.
A quick take by Bob Johnson, Gartner, is that “Heterogeneous integration with fan-out packaging gets you all the way to the 5nm silicon technology node through the year 2026.”
Future trends for advanced packaging are thought by Jan Vardaman, TechSearch International, to be these:Broadly, for AI, the consensus is that with the 10B+ transistor GPU products now available from NVIDIA, AMD, and others, along with the dedicated-architecture processors from Google and others, and with the concurrent availability of very high bandwidth memory in the form of stacked DRAM from Micron, SK Hynix, et al., we are just about to crack the barrier to seeing the widespread proliferation of commercially available products employing some useful form of Artificial Intelligence.
Terry Higashi, Tokyo Electron, in his keynote talk went so far as to assure us that, ten years hence, human beings and AI would be coming together in some kind of spectacular way to produce a kind of Singularity in which humans are empowered by AI, not enslaved by it.
Other SEMICON West talks in which AI was mentioned didn’t go quite so far out as Higashi-san, but the speakers did bring forth practical examples of how AI, aka Machine Learning, or Deep Learning, is making life easier in semiconductor manufacturing to the point where trained Machine Learning systems have created, for Micron Technology, a 25% improvement in time-to-yield ramps for new products and new fabrication technologies. According to Micron the 25% time-to-yield improvement, which they think is huge, is the result of only 3 years of effort with their Deep Learning project. Watch for more, they said, because “We are just getting started.”
Kathy Winter from Intel, in her keynote talk on “Big Data in Autonomous Driving,” brought in the same kind of thinking, which is that the huge amounts of data being generated by the embedded sensors (cameras, radar units, inertial measurement units, etc.) in ready-for-autonomous-operation vehicles can only be processed and interpreted by highly advanced silicon processors. These vehicles not only need to do their own heavy lifting for onboard data crunching and subsequent action; they also need to be in direct communication with vehicles around them, and the transportation grid as a whole, via the soon-to-be-rolled-out 5G wireless communication network.
It’s going to be a “Just let go of the wheel” future for us, which could be the start of the “Passenger Economy” or, at the least, a world in which “There’s no way it (autonomous driving) is not safer” than the distracted driving environment we as humans tolerate today.
We’re just getting started.
And then there’s China …
From the perspective of OSATs, Jan Vardaman presented this:
Dan Tracy, SEMI, has China in 2019 as “The key driver moving forward for semiconductor device fabrication and also for advanced packaging applications.”
Lung Chu, President, SEMI China, has it this way, with a deep-pockets Sino IC Fund of RMB 300B making strategic investments (please to note: not strategic subsidies) in the Middle Kingdom:
Word on the street at SEMICON West was that the China story is either going to be a boon for our industry, or that it will result in a debacle, just as China’s investment strategies in Solar PV, and in LEDs, have proven to be debacles for the supply chains in both those industries (albeit being boons to the consumers of those products!).
China and semiconductors: We’re just getting started.
CY2018 is expected to top the ’17 numbers by an additional ~$4B for the semiconductor capital equipment industry overall, thanks in no small part to investments by China, so, according to some of the Bulls and Bears panelists, the Age of Cycles isn’t expected to return until 2019, if return it ever does.
I’m going to take that to the bank … in my Autonomous Vehicle, of course.
We’re just getting started indeed.
From Santa Clara, CA, thanks for reading. ~PFW