Over the past two decades, the shift from fixed-function hardware to software-defined hardware has revolutionized industries ranging from networking to mobile communications. With software-defined hardware, developers can improve products and services by continually updating software rather than undertaking more costly and time-consuming hardware upgrades. Devices that were once rigid and task specific are now becoming programmable and flexible, allowing them to handle new tasks and demands.
Despite the functional benefits of software-defined hardware, its use was traditionally limited to industries that offered substantial unit volumes, such as smartphones, to amortize the fixed development costs. In these sectors, physically replacing or upgrading devices is much more expensive than making software updates. (See sidebar “Transforming industries through software-defined hardware” for more information on early use cases.) Industries that had lower hardware volumes did not feel the same urgency, because software development costs were in line with or exceeded those for device replacement. Today, however, AI is changing the cost-benefit balance by automating many routine software development tasks, thereby reducing the time and labor required and expanding the functional capabilities of the software.
Beyond improving efficiency, software-defined hardware may help companies win new customers and increase brand loyalty by improving device performance and enabling greater personalization. A car’s infotainment system, for instance, could provide customized entertainment and streaming options based on the driver’s previous choices. If the software-defined hardware incorporates AI or machine learning (ML) algorithms, it may take product performance to even greater heights and learn from interactions with customers.
As AI enables further product development cost reductions, even more industries, including aerospace, medical devices, and consumer devices, could accelerate scaling of software-defined hardware. First, however, they must update their organizational structures and operations to maximize AI’s benefits. Here’s a look at some changes that may help, with a focus on examples from the automotive industry.
Benefits of AI in software-defined hardware
For many companies, the advances in AI have come at an auspicious time. Software complexity has been increasing steadily and could rise even further. Consider the automotive sector: since 2021, the complexity of the average vehicle software platform and the total effort required to create it have both increased by about 40 percent annually (exhibit). Over the same period, however, software development productivity has increased only by about 6 percent per year.
Using AI to optimize software-defined hardware development
Advances in AI are expanding the toolbox of product developers and improving the development of software-defined hardware by automating, optimizing, and enhancing various stages of design, development, and testing.
AI-assisted design. Several generative AI (gen AI) tools can help engineers develop hardware descriptions and layouts, reducing manual effort. For instance, one tool optimizes hardware architecture design based on specified constraints and objectives. Other new generative design systems can explore a much larger universe of possible solutions than the previous generation of tools. By comparing the results of thousands of simulations, they can help identify a design that delivers the most favorable combination of attributes.
Software–hardware codevelopment. Using AI agents during product development can help bridge gaps between software and hardware design by ensuring consistent requirements across iterations. AI agents can also ensure that the hardware is easy to program and the software is optimized for hardware performance (for instance, AI can adjust software routines to improve utilization of GPUs, specialized accelerators, or other resources).
Hardware optimization. AI algorithms can optimize the allocation of resources, such as memory, logic blocks, and processing units, for software-defined hardware to meet objectives and enable adaptations if and when product requirements evolve. AI can also identify issues and flaws early in the design process through the use of simulations and improve product reliability and security. Hardware optimization can continue beyond the initial design stage because edge processing allows engineers to manage resources dynamically to meet objectives.
Accelerated testing. Deep-learning surrogates now allow engineers to replace many physical tests with faster virtual assessments that cost less.
With AI handling these tasks, more companies will be able to create software-defined hardware for the first time or accelerate innovation. London-based technology company Wayve, for example, is incorporating its LINGO-1 gen AI model into its self-driving car software to analyze driver inputs and data from vehicle sensors, including images captured on video. Using LINGO-1 capabilities, the vehicle’s software will then generate answers to common questions, such as “Why did you stop at this point?” Similarly, Waymo uses Carcraft, a software program that models how virtual cars navigate different on-road scenarios in actual cities, to improve autonomous driving. The virtual vehicles travel about eight million miles per day, much of this in complicated traffic situations, such as rotaries.
Using AI to create better end products and services
The benefits of software-defined hardware, enhanced by AI, translate into better products and services for end customers. As an example, compare the customer experience at electric vehicle (EV) start-ups with that of traditional OEMs. New EV OEMs, such as Tesla, which were not constrained by legacy hardware, have used software-defined hardware and delivered over-the-air (OTA) updates to improve vehicle performance, efficiency, and features beginning in 2018. That convenience is not possible with older, hardware-centric car designs. Other OEMs are also investing in software-defined hardware or forming joint ventures to create it.
Here are a few other ways that software-defined hardware and AI can improve end products and services. Note that these enhancements relate to products that companies sell to consumers or other businesses. Companies may also use software-defined hardware to improve their own operations (see sidebar “Making a warehouse more efficient”).
Optimal hardware performance and lifetime. AI can analyze data from devices, such as cars and medical equipment, to optimize software configurations for both efficiency and performance. For instance, one new EV OEM used AI to reduce the number of electronic control units in its vehicles from as many as 80 in some legacy systems to under ten.
Personalized user experiences. Gen AI can make software-defined systems more intelligent. In smart homes, for instance, gen AI systems could learn a household’s habits and automatically adjust everything from lighting to security. If the systems relied on traditional hardware, rather than software-defined hardware, such adjustments would be difficult and expensive to make.
Many medical devices are becoming increasingly software driven, with companies such as Medtronic creating AI-enhanced systems that adapt based on patient data, such as vital signs. In the consumer sector, devices such as Amazon Echo already rely on software to introduce new features, improve their intelligence, and enhance the user experience. As more companies begin incorporating AI into their products, they can use software-designed hardware to provide more personalized service. For instance, several new entrants have launched gen-AI-enabled digital personal assistants.
Better human–machine interfaces. Human–machine interfaces (HMIs) have long allowed people to interact with different devices, ranging from industrial machines to personal devices to service robots, but the communication channels can be frustrating or imperfect. Using AI, some companies are attempting to improve them. Within the automotive sector, for instance, the location technology company TomTom has developed an AI in-vehicle assistant that applies recent advances in AI to engage in complex, multifaceted conversations. The voice system proactively gives drivers useful information, rather than waiting for them to ask questions.
BMW’s Panoramic Vision is a new heads-up display that projects information, such as vehicle speed, onto windshields. It uses AI to track the driver’s eye movements to ensure that information always appears in a clear, easy-to-see location that will not interfere with the view of the road. The most important information appears in a darkened section at the base of the windshield, while less important data are projected slightly higher onto a clear section. The technology can automatically determine when certain info should be displayed in a priority spot. For instance, it prominently displays navigational information to help with parking when a driver is looking for or entering a spot.
Application of foundation models. Foundation models are trained on vast data sets and leverage advanced-machine-learning techniques to create better hardware. They have already been used to optimize performance and functionality in enterprise and consumer products. Embodied AI, which lets robots and other intelligent devices interact with their environment and learn from experience, could be the next important application. Some foundation-model-based robotics start-ups are hoping to create more generalized hardware with broader functionality that improves over time. (A new approach to model creation may help expedite development, as described in the sidebar “What are modular foundation models?”)
A new way of working
Harnessing AI’s power may require company-wide operational changes or industry-wide initiatives related to collaboration, market entry, decoupled hardware and software development, and data security and privacy.
Innovative collaborations
As more companies begin using AI to create software-defined hardware, they should ensure that hardware engineers and software developers collaborate from the program’s outset. They also should ensure, very early on, that the hardware design is flexible enough to allow software-designed hardware to make updates.
In some cases, companies may form nontraditional partnerships, such as with tech start-ups, to create software-defined hardware. They can facilitate such collaboration by mandating the use of standard protocols that allow dynamic management of network devices used in software-defined hardware, including the OpenFlow and the Network Configuration (NETCONF) protocols. While standard protocols can be useful in any collaboration, they are particularly important when two partners have never worked together and do not understand how the other company operates.
Some protocols are industry specific, including the Automotive Open System Architecture (AUTOSAR), which provides a robust framework for developing software-defined automotive hardware. AUTOSAR offers two complementary standards: Classic AUTOSAR, which is well suited for traditional embedded systems with strict real-time requirements, and Adaptive AUTOSAR, which is designed for more complex, dynamic, and software-intensive functions. Adaptive AUTOSAR enables the creation of scalable, modular, and secure applications, some of which may involve AI functionalities. For instance, Adaptive AUTOSAR is well suited to managing advanced-driver-assistance systems (ADAS), which may include AI-based perception and decision-making. Adaptive AUTOSAR also supports the deployment of applications that interact with radar, ultrasonic, and other sensors essential for ADAS, although specialized AI frameworks might still be required to handle other tasks, such as data processing.
Other protocols that may accelerate the development of software-defined hardware include the Safety-Oriented Architecture for Functional Safety in Edge Computing protocol, sponsored by the Scalable Open Architecture for Embedded Edge (SOAFEE) initiative, which focuses on creating safety standards for edge devices.
By following these common frameworks, developers can build flexible, scalable, and efficient systems that adapt to evolving technological demands.
As the universe of partners expands and diversifies, companies can encourage more AI-related innovation by lowering barriers to market entry. For instance, they could eliminate licensing fees for start-ups that want to develop hardware based on their technology. Overall, companies should focus on enabling potential partners, rather than controlling their activities.
Decoupled hardware and software development cycles
To create software-defined hardware, companies should decouple software and hardware development to enable greater flexibility and innovation. This does not mean that each element evolves with no input from the other group, since collaboration is more important than ever with software-defined hardware. It does mean, however, that hardware and software each follow their own timelines and update cycles. The pace set by the two groups will not necessarily be in sync.
Decoupling allows hardware engineers to design and optimize physical components independently, while software developers can create and update applications without being constrained by hardware limitations. By utilizing abstraction layers and standard interfaces, teams can iterate on software features and functionalities rapidly, fostering a more agile development process. Decoupling also facilitates easier integration of new technologies and upgrades, as hardware can evolve to meet changing demands without requiring a complete overhaul of the software stack. Ultimately, this approach enhances collaboration across disciplines and accelerates time to market for new products and solutions.
Legacy automotive OEMs are among those decoupling hardware and software development, partly because cars are becoming increasingly connected. For software, some OEMs outsource development to technology companies or innovative suppliers, while others try to build in-house capabilities. Internal software development is often costly and time-consuming, since companies typically need new tools, tech talent, and infrastructure.
More robust data security and privacy
Devices receiving real-time updates are vulnerable to cyberattacks. As companies venture further into the world of AI and software-defined hardware, they may want to separate safety-critical code from noncritical code, either physically or virtually, to protect users and preserve their privacy.
A long-term market outlook
If companies plan to update hardware continually so products can evolve, they must think ahead. What will consumers likely expect in ten years? What trends are gaining traction and could become standard in five years? As they plan for future upgrades, companies should consider not just potential demand but the competitive landscape and the costs of any improvements.
Leading companies are driving the shift toward software-defined hardware in more industries. AI will be pivotal in accelerating this shift and pushing the boundaries of what this technology can deliver. In addition to streamlining hardware and software development, AI can increase customer satisfaction and loyalty by enhancing the quality of end products and services. Eventually, the combination of AI and software-defined hardware may lead to a future in which devices continuously and instantly adapt to user needs.