Fundamental Shifts In 2018

What surprised the industry in 2018?  While business has been strong, markets are changing, product categories are shifting and clouds are forming on the horizon.

As 2018 comes to a close, most companies are pretty happy with the way everything turned out. Business has been booming, new product categories developing, and profits are meeting or beating market expectations. “2018 was indeed an exciting year for semiconductors, with growth in almost every sector and application,” notes Tom Wong, director of marketing for design IP at Cadence. “According to WSTS, worldwide semiconductor revenue will hit approximately $478 billion in 2018. This is an increase of 15.9% from 2017.”

New technologies are becoming mainstream. “2018 was surprisingly surprising,” says Thomas Uhrmann, director of business development at EV Group. “It was a strong year across a wide swath of technology. Optical sensing was very strong, but that was expected. The first major products are booming in the market now. Apple’s face ID had a large impact on this market. Customers are now forming consortia right and trying to prepare for the future.”

However, the warning signs on the horizon are growing. Capital spending is slowing, inventory is rising, and a trade war with China shows no sign of abatement. All of those can be highly damaging. Nobody wants to talk publicly about those possibilities yet, but they are beginning to tilt the outlook for 2019.

Artificial intelligence

Meanwhile, in 2018, artificial Intelligence (AI) is perhaps the buzzword for the year. Lauro Rizzatti, a verification consultant agrees. “Artificial intelligence, machine learning (ML) and deep learning became the industry’s words of the year in 2018 and created intriguing areas where startups are making inroads in innovative ways. Neural network acceleration, for example, got a lift from a software stack that sits atop an FPGA, optimizing its performance and concealing the FPGA programming from the user to ease deployment.”

Nobody gets an award for having seen this trend. “We were right that AI would rapidly expand to consume all industry sectors, but this wasn’t a hard one to predict—AI is ‘the’ hot topic in technology,” says Simon Forrest, director of Connectivity & Connected Home at Imagination Technologies. “Cloud AI has certainly made strides in 2018, but many companies also used AI in name alone as a massive marketing tool. Many are not using AI in the truest sense of the word. The majority are exploiting pattern-matching algorithms alongside big data analytics, then claiming it as AI. This resulted in ‘IoT’ largely being replaced by ‘smart technology’, with that smartness implying some form of AI.”

But many people at least got a piece of this right. “We said that machine learning (ML) would present many new challenges for systems, semiconductor and EDA suppliers,” said Sergio Marchese, technical marketing manager at OneSpin Solutions. “Specific to EDA, we predicted that ML and AI would lead to an evolution in design practices and development tools. We are still in the early phases of this transition, but some changes are already clear. The statistical nature of ML means that bugs are more data-driven, leading to a new emphasis on verification of datapaths in addition to control logic. The floating-point unit formal app that we recently introduced is one example of this evolution.”

While AI exploded, another area contracted. “In 2018, we saw the slowdown in bitcoin mining and its impact on the foundry and crypto SoC business,” says Wong. “But don’t worry, the industry is very resilient and is already moving to the next big thing—ML/AI SoCs. Just look at the VC activities in the U.S. and China and you will get the picture.”

The rise in development of special accelerator chips for the Cloud is also interesting. These chips defy the rules in some sense because they are being designed and built to sell services offered by the same company that is developing them.

Disappointments
But not everything was great in 2018. “It’s fair to say that augmented reality (AR) didn’t really move forward much in 2018,” says Forrest. “Mass market adoption of AR glasses depends on the need to be reasonably priced. The glasses themselves also need to be sleeker in design, while issues with battery life still need to be addressed. The potential is certainly there for AR, but its success is unfortunately tied into the success of wearable displays in general and will get thwarted somewhat in 2019 by the continued disillusionment with virtual reality (VR).”

Wong agrees. “AR/VR did not quite make it to the party. While a few high-end phones support AR/VR, most of them are still a work in progress.”

Everyone knows that mobile phones are no longer the growth leader, but 2018 was the wakeup call on that front. “I had predicted smartphone shipments would reach 1.6 billion units in 2018, compared to worldwide shipments of 1.472 billion units in 2017,” says Wong. “Unfortunately, the market was not that kind. Right now, it appears the smartphone market will close 2018 with worldwide shipments of about 1.5 billion units. I thought we would have some growth in 2018, but worldwide unit shipments were flat. This is clear confirmation that the worldwide smartphone market is saturated. Even in China, unit sales in 2018 came in below unit shipments in 2017. We also witnessed a change in ranking for top-tier suppliers and saw a strong showing by Vivo, OPPO and Xiaomi in China. All of these mid-priced Chinese smartphones have really good industrial designs and premium features. I had predicted Xiaomi would go public in 2018 with a valuation of $60 billion (USD). Well, Xiaomi did go public on the Hong Kong Stock Exchange during the summer of 2018, but at a valuation of $54 billion.”

Foundry changes
2018 brought major change to foundries, as well. “The most interesting news this year is the announcement by GlobalFoundries that they are stop development of 7nm and any advanced node after 7nm,” says Navraj Nandra, senior director of product marketing for the Solutions Group at Synopsys. “In the past there was always Intel, GF, Samsung, TSMC, with fairly well mapped out development paths. GF has said it is expensive and that they were not seeing the return.”

The industry has been talking about an increasing percentage of designs staying on older nodes. “GlobalFoundries has seen a sweet spot for their business at 14nm and 28nm/22nm, especially with FD-SOI,” adds Nandra. “They have found a niche in specialty technologies, and the mainstream finFET node right now is 16/14/12. GF is a substantial provider at that node. Analog and RF will become their focus. The opportunity is rich with all of the interest in IoT (edge or Industrial) that use these types of technologies. Automotive – there are customers wanting 28nm, and FD-SOI is interesting here.”

Competition is heating up on older nodes, as well, as both demand and capacity continue to grow. “China is still behind in semiconductors,” says EV Group’s Uhrmann. “A lot of fabs are being built and a lot of 300mm capacity is coming online. We are seeing China acquire a lot of technologies. People are getting worried about a trade war, and so are reconsidering joint ventures. We already see a lot of our customers re-concentrating efforts back to the U.S.”

With foundries revamping old nodes, the rate at which new processes are being released is accelerating. Many of these are focused on particular industries, such as IoT or automotive. This is putting a strain on the IP industry.

“In the past you were developing one USB that covered multiple market segments or one DDR,” explains Nandra. “You now have a dedicated USB 2 for IoT and one for consumer of mobile. And if you look at how the IP is designed, even though the electrical specifications are the same, the actual layout looks different. It is no longer a certainty that if you build a piece of IP on a certain process that you will get a lot of customers.”

The IP industry itself is being shaken up by the introduction of the RISC-V open source instruction-set architecture (ISA). “I tend to underestimate how big momentum is,” says Krste Asanovic, professor at UC Berkeley and chief architect for SiFive. “Even though I am enthusiastic and believe that it is doing really well, it always seems to exceed expectations—and that continues to be true. The degree of interest and upswell is amazing.”

The RISC-V ecosystem is still behind rival ISAs when it comes to commercial offerings, but the gap is shrinking. “Processor ecosystems did prove very important, perhaps even a bit more prominent, than I had thought,” says Frank Schirrmeister, senior group director, product management and marketing for Cadence. “Qualcomm has now disclosed that it will be shipping RISC-V in a high-volume product in 2019. With others like Western Digital and NVIDIA already on the RISC-V train, joined by several commercial IP providers, the area of processor ecosystems has been interesting to watch. Alliances are developing fast, and friends on one day may be foes the next.”

“Some of the big milestones this year were pushing things through ratification,” says Asanovic. “The memory model spec was ratified, the debug spec was ratified, the base ISA spec, the foundation is going through ratification. At the end of 2018, all of the base should be in place. This is new. It is the entire community coming together to agree on an ISA standard. This has never happened before.”

Asanovic expected more competition in one area. “One surprise is with 64-bit embedded. We see a lot of demand, but not much competition. There are few competing solutions, especially for markets such as storage. Companies there are moving to 64-bit because it makes sense when you have large storage address spaces, but there are few solutions available.”

Another standard that made the news in 2018 was the long-awaited Portable Stimulus Standard. “As anticipated, the Portable Test and Stimulus standard was approved by the Accellera Board of Directors and moved into the public domain, a great leap forward for all chip design verification engineers,” noted Michiel Ligthart, president and COO for Verific Design Automation. “However, based on our experience with UPF, it may still take a while before it finds widespread use in the industry.”

Adoption rate depends upon the severity of the need. “The adoption rate of PSS, as well as its critical applications, is having ramifications across system-level infrastructure testing on emulation, cache coherency, automotive testbenches and large communications devices,” says Dave Kelf, CMO for Breker Verification Systems. “As we suggested in our prediction, building machine learning into EDA tools not only can encapsulate the knowledge of leading verification engineers, but can allow EDA approaches to evolve to more effectively target these kinds of applications. For example, understanding coverage, accelerating debug and producing more effective corner case tests are all aided with this technology.”

Bob Smith, executive director for ESD Alliance, a SEMI Strategic Association Partner, also sees other changes becoming important. “2018 marked the beginning of a new era for collaboration between design and manufacturing, as it has become obvious that decisions made during design can have a significant effect on what happens during manufacturing. Because of this, design and manufacturing are drawing closer together by necessity. In recognition of the design ecosystem’s importance in the electronic product supply chain, the ESD Alliance aligned itself with SEMI and is now a SEMI Strategic Association Partner.”

Design and verification challenges
Automotive is one of the growing application areas, but it has some unique challenges. “Functional safety has become a huge topic especially in automotive, but it extends into other areas like medical and aero/defense application domains very fast,” says Schirrmeister. “Classic ‘functional verification planning’ is joined by ‘safety verification planning’ with all the aspects around FMEDA (failure modes, effects, and diagnostic analysis), and next up is ‘security verification planning’. The key aspect to remember here, likely a key trend going forward, are threat models at several levels of abstraction—from RTL to system and software level—representing as early as possible the path that hackers could use to attack chip and system security, allowing verification to check proper counter measures.”

2018 was the year of the processor vulnerability. “Two notable attacks, Meltdown and Spectre, affected nearly every modern CPU used in the last 20 years,” notes Martin Scott, CTO and senior vice president/general manager Rambus’ Cryptography Research Division. “As some security experts predicted, these attacks had one thing in common—they exploited complexity and the resulting imperfections in security design inherent in building high-performance mainstream processors. This past year demonstrated that, more than ever, a best security practice is the isolation of the highest level of security operation from general-purpose compute optimized for performance. Rambus’ RISC-V based security co-processor, for example, was designed by security experts to be secure by design and to be resilient to attacks that exploit the vulnerabilities that became more visible this past year.”

The IoT is also very concerned with security. “We started to see virtualization in some IoT devices this year to help improve device security overall,” says Imagination’s Forrest. “But 2018 hasn’t delivered any spectacular failures in IoT security—nothing that hasn’t been experienced already or previously forewarned. Instead, we’ve seen the continued mantra of ‘buyer beware,’ with IoT security warnings only slightly heeded by consumers. Hackers with serious intent tend to go after the cloud-based infrastructure rather than individual IoT end devices, because there is more to be gained by taking a service offline or compromising systems holding customer data.”

Home IoT also appears to be accelerating. “On the IoT side my predictions came true, although we’ll need a little more time for the packaging technology predictions to come true,” says Marc Greenberg, group director of product marketing for Cadence’s IP Group. “Over the course of this year, I personally invested in a connected security camera system for my home. This technology is becoming pervasive. Security devices are getting smarter with AI. These cameras come with an optional server-based AI service that will help to filter out alerts due to normal activity. The digital assistant portal devices also have become pervasive over the course of the year. I see lots of sales on connected home devices, particularly lighting/socket control devices, in the sub-$20 range.”

These edge devices are spurring a lot of innovation. “Edge computing may well be 2018’s tech term of the year and has spawned a flurry of design activities,” says Schirrmeister. “The intensity with which activities have spawned in this domain, for instance at ArmTechCon, certainly surprised me.”

And the drive to put AI on the edge is causing new design architectures. “There has been a rethink on compute architectures for cache coherency, for heterogenous computing,” points out Synopsys’ Nandra. “In 2018, different approaches have been attempted to solve ML inference challenge. They are all trying to work out how to do inferencing on an edge device and being able to quickly process data so that they don’t have to upload information from the Cloud and to do things in real time. That has opened a big debate about von Neumann compute architectures to different approaches where you are separating memory, accelerator chips—be they dedicated FPGAs, dedicated GPUs, application specific processors—and new chips that have blocks that talk with each other.”

2018 was a foundational year. “That includes the beginning of the 5G era, better prioritizing safety and driver trust in the design of autonomous computing, and introducing the infrastructure needed in a world of one trillion connected devices,” says Nandan Nayampally, vice president and general manager of Arm’s Client Line of Business. “This year also proved the pace of innovation that transformed smartphones into the compute platform of choice is now powering and transforming the larger screen productivity devices. The question now for the broader PC industry is, ‘Is it ready to liberate itself from the slowing pace of Moore’s law and deliver the mobile productivity experience consumers and businesses will need as we enter the 5G era?’”

Conclusion
The semiconductor industry is changing and evolving rapidly. While mobile phones continue to provide a solid base, innovation is happening in several other areas. The entire ecosystem is having to re-evaluate decisions that were made decades ago, and design innovation is becoming more important than it has ever been. The robustness of 2018 has given companies the opportunity to change, but 2019 is looking a lot cloudier. Those who have managed to innovate and change their practices may be set to pull ahead of the pack.

Related Stories

Big Changes For Mainstream Chip Architectures

Some Chipmakers Sidestep Scaling, Others Hedge

AI Market Ramps Everywhere