What you'll learn:
- How to reduce AI's energy burden, which will grow exponentially as AI becomes more widely adopted.
- How gallium nitride is essential to mitigating AI's power consumption and reducing emissions.
- The importance of third-generation semiconductors in enabling IoT devices and EVs to scale.
The entire technology sector is undergoing massive transformations and shifts as it works to meet the demands of artificial intelligence (AI). While AI isn’t a new technology, there’s no denying it’s become a household term over the last 12 months due to the emergence of generative AI and consumer-focused applications, such as chatbots.
As with any technology releasing a "killer app" that grabs consumer attention, it’s important to avoid getting caught up in the hype if one wants to understand the underlying value proposition and corresponding technology needs and requirements. For AI, as is the case with many technologies, its progression will be limited by different "bottlenecks" depending on maturity. In other words, the challenges of AI today are likely to be different than the challenges of tomorrow.
This article focuses on AI and the connection to advanced materials such as compound semiconductors (Fig. 1).
Up to this point, AI has largely relied on scaling or incremental improvement of legacy technology. Case in point: The acceleration of AI in 2023 was largely driven by a convergence of massive amounts of data harvested from things like social media, developments in machine learning, and hardware enablers such as edge computing. As AI moves forward, simple scaling or incremental improvement will be insufficient to meet market demands, and "new" technology will be required.
One key trend that will drive AI of the future will be to transform our "smart" devices and make them "intelligent." This is best explained by an example. Today, in the world of the Internet of Things (IoT), it’s routine to control various aspects of our homes and vehicles via our mobile phones. For example, we can turn up the thermostat at home or turn off the light in the bedroom using our mobile phone while sitting at work miles away. Such technology is "smart."
Our devices are making it more convenient to carry out simple, routine operations remotely. The next step would be to make this scenario "intelligent." For example, an intelligent home might detect that I’m nearing my residence on my commute from work and determine that lately I want the thermostat set to a given value and make the adjustment. When I arrive, the home may confirm that it is indeed me and not my partner, since we have different temperature preferences.
In this scenario, my home is making decisions, not simply executing my instructions, and therefore, the "smart" device converges with AI to create "dynamic AI." In the future, there will be an insatiable demand to make everything "intelligent."
The Intelligent World Fostered by AI
The AI of the intelligent world requires more than computational power and will be enabled by other technologies. Most of these technologies rely on new materials, particularly compound semiconductors.
AI in the intelligent world has five key processes:
- Detection: AI must operate on something. So far, it’s operating on data that’s built up by a proliferation of such things as social media. However, in the future, we’ll need to operate on more diverse data types. As with any data-driven technology, the quality of the output depends on the input. Therefore, to realize the full potential of AI for, say, driverless cars and smart healthcare, high-fidelity input data is required. Most of this input data will be generated through some sort of machine-driven sensing. For example, AI in driverless vehicles will require that a vehicle optically image its environment in ultra-high resolution.
- Data transfer for computation: Once something is detected, the information must be transferred to a computational device. Such transfer needs to move massive amounts of data rapidly with low latency and extremely high reliability ("zero" error rate).
- Data processing/computation: To date, this has been the focus for AI, with AI algorithms being rapidly advanced. Continuous evolution in computing hardware and things as simple as huge amounts of cheap memory have enabled computational power to rise to the challenge posed by the first wave of AI. The challenge will only get steeper. This is where technologies like quantum computing at scale enter the picture.
- Data transfer for output: This is the reverse of number 2, with the same stringent criteria applied.
- Output: In nearly all cases, AI will be expected to output something. In many cases, this will require that information be displayed. Often, the requirement will be displayed on a small-form-factor output or in some sort of immersive environment (e.g., AR/VR). The quality of these displays is critical to make the entire system effective.
As noted above, the current generation of AI has focused on number 3, and this is likely to continue with the key advances required, as discussed below. However, the power of AI will have other roadblocks unrelated to computational power. Let's look deeper into the other four areas (actually three, since there are two data-transfer processes).
AI Intelligent World Process: Detection
For the intelligent world to have a chance, advanced sensing is a must. Technologies such as 3D facial recognition will be a basic requirement in this world. Such imaging will extend to larger distances (e.g., LiDAR in cars). Essentially, detection will expand beyond what can be "seen," leading to exciting advances in medical care where there will be a step function improvement in early detection—e.g., our devices will continuously monitor (non-invasively) our blood and look for biomarkers.
The above-mentioned forms of detection have a common thread. They require light of a specific wavelength to interact with the environment and provide an output signal that’s the input for AI.
Such optoelectronic sensors rely on compound semiconductors. These have developed over many years, starting with telecom data transceivers and culminating in advanced devices like vertical cavity surface emitting lasers (VCSELs), which underpin 3D sensing.
Development in compound semiconductors for sensing is very active and will continue to focus on scaling technology that operates over a large spectrum (from deep UV to LW IR). Advances in this area couple fundamental materials development with volume semiconductor manufacturing. The engineering and enablement of these materials require control on the atomic scale using a technique known as epitaxy. To date, the leader in this area is IQE, whose current product roadmap is aligned with the intelligent world of the future.
AI Intelligent World Process: Data Transfer
As noted, there are two key data-transfer steps for dynamic AI. We’ll lump them together since the requirements are the same. Specifically, dynamic AI relies on the ability to move massive amounts of data rapidly with low latency and a "zero" error rate.
Again, this is the realm of compound semiconductors that underpin technologies such as rapid data communications (e.g., transceivers discussed above) and 5G wireless communication. Compound semiconductor materials such as gallium nitride (GaN) are essential in today's and tomorrow's communications infrastructure (e.g., base stations). On the mobile device side, gallium-arsenide (GaAs)-based HBTs are the backbone of front-end modules for 5G handsets.
Going forward, it’s not far-fetched to assume that all smart components of the IoT will need to function as a mobile phone, from our cars to the streetlights, to our refrigerators, to the very clothes we wear. As the amount of data scales, the need for bandwidth and speed will scale. This will move things to higher frequencies and into a new realm known as FR3. At these frequencies, compound semiconductor materials offer excellent performance and efficiency, and they will continue to be driven to improve.
As with the photonic materials discussed in the prior section, the stringent specifications imposed by the needs of AI require materials engineering and control on the atomic scale, again requiring epitaxy.
AI Intelligent World Process: Output
One of the most common images associated with future AI is the immersive augmented-reality/virtual-reality (AR/VR) environment. This will require ultra-high-resolution displays that are powered by extremely small (defined size) light-emitting diodes or micro-LEDs. Again, compound semiconductor materials are the only game in town for micro-LEDs, and epitaxy is a requirement for the creation of leading-edge materials. In response to this need, IQE is developing a portfolio of red, blue, and green micro-LED emitters based on both GaN and GaAs.
As can be seen, the dynamic AI world depends heavily on compound semiconductors. Current generations already enable the basic capability needed, and next-gen advances will drive future generations of AI.
Compute Advances to Meet Quantum-Computing Speeds
Although there are other key components to dynamic AI, we can’t forget that the development of these components will occur concurrently with further development of the computing process. As mentioned above, the speed and computational requirements of dynamic AI are driving the advances of a new type of computer—the quantum computer. Although still in its infancy with many challenges on the horizon, the quantum computer holds promise and may be the only way to fully realize the potential of the dynamic AI world.
For many, the quantum computer conjures up images of science fiction, and this is for good reason. Moreover, it will require a whole new set of materials (e.g., superconductors, oxides, 2D materials, etc.).
Although different than compound semiconductors, innovation in this area is still a materials science/engineering problem. It’s expected that fine control enabled by such techniques as epitaxy will be required to advance the quantum computer to a maturity that meets the needs of dynamic AI. Companies such as IQE, which has a pedigree for atomic-level manipulation of complex semiconductors, will be heavily involved in this area as it progresses (Fig. 2).
Handling Power Consumption with Compound Semiconductors
Up to this point, we’ve ignored a proverbial elephant in the room. As dynamic AI progresses to monitor our entire world, move massive amounts of data, and computationally process information, often with a high-resolution output display, the energy consumption of all components in the system will scale. As such, we hit a roadblock that’s somewhat ironic. To create a world enabled by dynamic AI, we are exponentially worsening the global environmental crisis.
There is hope, though. In addition to their superior optical and electrical characteristics, compound semiconductors have one other key property, arguably their most important characteristic. They’re extremely efficient.
A real-world example that brings this to life comes from the lighting industry. Most people are familiar with LED light bulbs/fixtures that have largely replaced filament bulbs. LED bulbs are made with the compound semiconductor GaN. Comparing an LED bulb with its filament counterpart, the LED bulb uses ~10% of the electricity as the filament bulb uses for the same light output.
This marked difference is due to the efficiency of GaN in turning input electricity into output light. The same occurs for highly efficient power supplies (for mobile devices) based on GaN. Significantly less input energy is wasted as heat, resulting in a much more efficient system.
As AI progresses, GaN will feature heavily in power components to avoid energy consumption scaling with dynamic AI progression.
Creating the Dynamic AI World of Tomorrow
The recent advances in AI have captured the world's attention, and many exciting developments are looming with AI still in its infancy. AI will drive us from the "smart" world to the "intelligent" world. With this shift, other capabilities outside of AI computation (today's focus) rise in importance.
This drive to the intelligent world will mean that the capability of semiconductor materials, underpinned by compound semiconductors and novel material epitaxy, will come to the fore and become a key enabler of the dynamic AI world of tomorrow.