Semiconductors and Artificial Intelligence

Artificial intelligence (AI) applications are everywhere, from big data analytics and military equipment to facial recognition software and self-driving cars. And they bring new challenges and opportunities to the semiconductor industry every day.

As a reminder, AI describes a machine or software application’s ability to reason, learn, and act in a manner similar to human cognition. In essence, AI makes it possible for machines to think.

The beginnings of AI date back to the 1950s, but recent advances in AI technology have seen a renaissance in the field. The development of machine-learning algorithms capable of processing massive amounts of data has opened new possibilities for AI devices. Today’s AI applications can not only process data but also learn from experience and apply that experience to improve how they function.

With AI applications gaining traction in the industrial, retail, health care, military, research, and consumer sectors, demand for specialized sensors, integrated circuits, improved memory, and enhanced processors is increasing. And this demand is changing the semiconductor supply chain by directly impacting design and manufacturing decisions.

 

How will AI affect semiconductor design and production?

AI demands will have lasting impacts on semiconductor design and production. In large part, this is because the amount of data processed and stored by AI applications is massive.

Semiconductor architectural improvements are needed to address data use in AI-integrated circuits. Improvements in semiconductor design for AI will be less about improving overall performance and more about speeding the movement of data in and out of memory with increased power and more efficient memory systems.

One option is the design of chips for AI neural networks that perform like human brain synapses. Instead of sending constant signals, such chips would “fire” and send data only when needed.

Nonvolatile memory may also see more use in AI-related semiconductor designs. Nonvolatile memory can hold saved data without power. Combining nonvolatile memory on chips with processing logic would make “system on a chip” processors possible, which could meet the demands of AI algorithms.

While semiconductor design improvements are emerging to meet the data demands of AI applications, they pose potential production challenges. As a result of memory needs, AI chips today are quite large. With this large chip size, it is not economically easy for a chip vendor to make money while working on a specialized hardware. This is because it is very costly to manufacture a specialized AI chip for every application.

A general-purpose AI platform would help address this challenge. System and chip vendors would still be able to augment the general-purpose platform with accelerators, sensors, and inputs/outputs. This would allow manufacturers to customize the platform for the different workload requirements of any application while also saving on costs. An additional benefit of a general-purpose AI platform is that it can facilitate faster evolution of an application ecosystem.

From a production standpoint, the semiconductor industry will also itself benefit from AI adoption. AI will be present at all process points, proving the data needed to reduce material losses, improve production efficiency, and reduce production times.

Read more about semiconductors and artificial intelligence in the IRDS™ Roadmap

Access the 2020 IRDS™ Roadmap

 

Why semiconductor companies must define their AI strategy

The semiconductor market has, for most of the last decade, seen much of its profits tied to the smartphone and mobile device market. As the smartphone market begins to plateau, the semiconductor industry must find other growth opportunities.

AI applications, especially in the big data, autonomous vehicles, and industrial robotics industries, can provide those opportunities. By defining and then putting together their AI strategies now, semiconductor manufacturers can position themselves to take full advantage of the spreading AI market.

 

How will AI affect the semiconductor market?

AI offers semiconductor companies the chance to get the most value from the technology stack, the collection of hardware and services used to run applications. In the software-dependent world of PCs and mobile devices, the semiconductor industry is only able to capture 20 to 30 percent of the total value of the PC stack and as little as 10 to 20 percent of the mobile market.

Within the AI sector, the technology stack requires more hardware, especially in the fields of memory and sensors. This may allow the semiconductor market to control 40 to 50 percent of the total value of the stack, according to the Redline Group.

In addition, many AI applications will require specialized end-to-end solutions, which will necessitate changes to the semiconductor supply chain. Semiconductor companies—especially smaller companies producing niche products for the automotive and IoT industries—will be able to capitalize on markets by providing customized microvertical solutions addressing customer pain points related to storage, memory, and specialized computing needs.

How does AI change the demand for semiconductor chips?

The global AI market is forecast to grow to $390.9 billion by 2025, representing a compound annual growth rate of 55.6 percent over that short period. Hardware lies at the foundation of each AI application.

Storage will see the highest growth, but the semiconductor industry will reap the most profit by supplying computing, memory, and networking solutions. Demand for semiconductor chips will mirror the rapid ascent of the AI market.

Read more about semiconductors and artificial intelligence in the IRDS™ Roadmap

Access the 2020 IRDS™ Roadmap

 

How AI technology provides opportunities for semiconductor companies

According to McKinsey, AI accelerator chips (chips designed to work with neural networks and machine learning) will see a growth rate of approximately 18 percent annually—five times greater than that seen for semiconductors used in non-AI applications. Areas of high growth will include AI chips for autonomous vehicles and in the broader field of neural networks.

Neural networks are specialized AI algorithms based on the human brain. The networks are capable of interpreting sensory data and delivering patterns in large amounts of unstructured data. Neural networks find use in predictive analysis, facial recognition, targeted marketing, and self-driving cars. And they require AI accelerators and multiple inferencing chips, all of which the semiconductor industry will supply.

How can semiconductor companies benefit from AI technology?

AI adoption holds the possibility for growth in the following areas of semiconductor manufacturing:

  • Workload-specific AI accelerators
  • Nonvolatile memory
  • High-speed interconnected hardware
  • High-bandwidth memory
  • On-chip memory
  • Storage
  • Networking chips

Investing in research and development while building relationships with AI software providers will help chip manufacturers capture their share of these markets—if they can meet the coming demand.

Read more about semiconductors and artificial intelligence in the IRDS™ Roadmap

Access the 2020 IRDS™ Roadmap

 

Impact of artificial intelligence on the semiconductor industry

The immediate future of AI has the potential to put strain on the industry supply chain unless semiconductor manufacturers plan to meet demand now. At the same time, the industry will itself benefit from AI, whose applications throughout the manufacturing process will improve efficiency while cutting costs.

How will AI technology affect semiconductor production?

Just as other industries are embracing AI, so too is the semiconductor industry (PDF, 10 MB). AI expertise coupled with high-performance computing will allow manufacturers to develop new efficiency benchmarks and increase output.

One of the key challenges to the semiconductor supply chain is chip production processing time. The time between initial processing and the final product takes weeks. And during this time, up to 30 percent of production costs is lost to testing and yield losses.

Embedding AI applications into the production cycle allows companies to systematically analyze losses at every stage of production so manufacturers can optimize operating processes. This ability will become even more valuable when working with next-generation semiconductor materials, which tend to be more expensive (and volatile) than traditional silicon.

How will AI technology affect the workforce in the semiconductor industry?

While the rise of AI brings many opportunities to the semiconductor industry, it also heralds a crisis in talent acquisition. The larger tech companies—most notably Google, Apple, Facebook, Amazon, and the like—are investing heavily in AI research, development, and implementation, especially in the arenas of big data analytics and deep learning.

This represents two challenges to chip makers. First, the major players in the AI industry increasingly develop their own hardware as this allows them to customize proprietary hardware to match their AI applications’ specific needs. This move toward in-house chip production, by extension, means the largest tech companies will purchase less from dedicated chip manufacturers.

Second—and this is where workforce considerations come into play—tech giants designing and manufacturing their own chips in house will need employees. With limited talent pools in both AI and the semiconductor industry, this will lead to talent shortages.

Read more about semiconductors and artificial intelligence in the IRDS™ Roadmap

Access the 2020 IRDS™ Roadmap

 

Future of semiconductors and artificial intelligence

Self-driving cars. High-performance computing. Quantum computing. AI makes what was science fiction at the turn of the century into reality. With these AI advances come demands for new semiconductor technology and deep changes to the industry.

How is artificial intelligence expected to affect the semiconductor industry in the future?

To adapt to an industry increasingly dominated by the need for AI hardware, semiconductor manufacturers will need to provide industry-specific end-to-end solutions, innovation, and the development of new software ecosystems.

End-to-end services will require chip makers to work with partners to develop industry-specific AI hardware. While this may limit the semiconductor manufacturer to working with only certain industries, the alternative—the traditional production of general products—may not attract the same customers it does at present. An exception would be the production of cross-industry solutions that serve the needs of an interrelated group of industries.

With the production of specialized products comes the need to develop existing ecosystems with partners and software developers. The goal of such ecosystems is to develop relationships in which partners rely on and prefer the semiconductor company’s hardware. Semiconductor manufacturers will need to produce hardware that partners cannot find elsewhere at similar value. Such hardware—coupled with simple interfaces, dev kits, and excellent technical support—will help build long-lasting relationships with AI developers.

Innovation, as always, plays a role in the future of semiconductors. In addition to the ongoing efforts to circumvent the limitations of Moore’s Law, semiconductor research and development will need to consider how sensors, memory, and microprocessors enable and support emerging AI applications. Focusing on serving the needs of AI and the equally important IoT industry will help keep chip makers at the forefront of the industry.

What is driving the popularity of artificial intelligence in the semiconductor industry? 

Demand from both the public and private sectors is driving the rapid development of AI—and as a result the importance of AI to the semiconductor industry. Of special note is the trend toward advanced driver assistance systems and electric vehicles. Even if the arrival of truly autonomous vehicles in large numbers remains years away, automotive AI applications for monitoring engine performance, mileage, and driver habits are already here. Insurance companies are already using in-car AI apps to evaluate driving habits and determine premium rates.

While the smartphone industry is plateauing in terms of growth, the demand for embedded AI in mobile devices is growing. Phones use AI for navigation, for voice-to-text software, for facial recognition security, and for personal assistants. The advent of Alexa and other smart home hubs—and their ability to be controlled from afar by phone apps—represents another growth area for AI.

Then there are the uses for AI the general public is only tangentially aware of. City planners increasingly rely on AI to report on traffic volume, sewer usage, and infrastructure maintenance. Utility companies use AI to set electricity and water rates or to alert technicians to incidents or maintenance events.

Retail and online retail stores use AI to predict consumer needs and preferences—with what some see as alarming precision. Similar software is used by major social network platforms when choosing content and ads for individual users. AI has applications in health care, bioscience, industry, government, and the military—anywhere where large amounts of data need to be processed quickly, analyzed, and acted upon.

Why must the semiconductor industry embrace artificial intelligence?

AI has negative uses as well as positive—the Cambridge Analytica (PDF, 423 KB) scandal proved how powerful a tool AI can be when used to identify and manipulate people’s behavior and opinions. Like so many technologies, AI is a double-edged sword. One thing it is not is temporary. AI applications are here to stay and will only become more commonplace and complicated with time. As every AI task needs to be founded on reliable hardware, the semiconductor industry has a vested interest in seeing AI succeed.

Interested in learning more about semiconductors and AI? Consider reading the International Roadmap for Devices and Systems (IRDS™). The IRDS™ is a set of predictions that examine the future of the electronics, semiconductor, and computer industries over a fifteen-year horizon. It encompasses a number of critical domains and technologies, from application needs down through devices and manufacturing. Join the IRDS™ Technical Community to download the roadmap and stay informed of our latest activities.

How to Download IRDS™

Access the 2020 IRDS™ Roadmap