As robots become more intelligent, autonomous and connected, much of the attention naturally falls on increasingly powerful AI models.
But underneath those models is an increasingly complex collection of sensors, processors, control systems and security hardware that must operate reliably – often within strict limits on power consumption, size and heat.
Lattice Semiconductor is one of the companies developing technology for this less visible layer of the robotics stack. Founded in 1983 and headquartered in Oregon, the company specializes in low-power field programmable gate arrays (FPGAs), which can be programmed and reprogrammed for functions ranging from machine vision and sensor fusion to motor control, industrial networking and hardware security.
The flexibility of FPGAs is becoming particularly relevant as robotics moves toward greater edge processing. Lattice is targeting applications including industrial robots, autonomous mobile robots and other physical AI systems, where deterministic real-time control, low latency and power efficiency can be as important as raw computing performance.
In this interview, Robotics & Automation News speaks with Karl Wachswender, senior principal system architect, industrial at Lattice Semiconductor, about how programmable hardware is evolving alongside the robotics industry.
Wachswender discusses why more perception, sensor fusion, object tracking and motor-control workloads are moving onto robots rather than relying on cloud processing, and how programmable architectures can allow machines to evolve after deployment instead of becoming obsolete as requirements change.
He also considers an area that receives considerably less attention than AI models themselves: the underlying infrastructure required to provide the right data at the right time.
Security is another central theme. As robots become connected to enterprise networks and operate increasingly close to people, Wachswender argues that security needs to begin at the hardware level, with FPGAs potentially providing a hardware root of trust from the moment a system powers on.
The discussion also looks ahead to the semiconductor opportunities created by humanoids, industrial automation, autonomous mobile robots and healthcare robotics as intelligent machines become more widely deployed.
Interview with Karl Wachswender
Robotics & Automation News: Humanoid robots are becoming increasingly capable, but they also have strict power, size and thermal constraints. How are low-power programmable devices changing the way robotics companies design their next generation of intelligent machines?
Karl Wachswender: Because they are so versatile, low power field programmable gate arrays (FPGAs) are allowing humanoid robotics developers to take a more holistic approach to system design.
A single FPGA can support functions like power gating, protocol bridging, and hardware security simultaneously, which opens the door for true, end-to-end optimization in a way that was previously unavailable in these inherently resource-limited environments.
This kind of parallel processing power helps support the deployment of more powerful AI models at the edge, which, in turn, support more intelligent robots.
When robots can sense, process, and act upon environmental data in real time, all without straining the central processing unit, they can be trusted with more autonomy.
R&AN: Edge AI is moving rapidly from cloud-connected demonstrations to real-time decision-making on robots themselves. Which AI workloads do you believe will increasingly move onto the robot, and which will continue to rely on cloud computing?
KW: When deciding what needs to be executed at the edge versus in the cloud, it’s less about the workload itself and more about the intended outcome.
How quickly do these tasks need to occur, and what’ll happen if they’re subject to any increases in latency? Will putting the onus on edge components provide a net benefit, or will it overwhelm their limited capacity?
Workloads tied to real-time perception and motor control (think depth processing, sensor fusion, and object tracking) will be edge-bound, since they require sub-microsecond determinism that roundtrips to the cloud can’t provide. Less time-sensitive processes, like enterprise analytics tasks, can continue to happen in the cloud.
This kind of edge vs. cloud prioritization can help balance urgency with caution, enabling the quick execution of time-sensitive workloads without tipping the scale too far towards the edge.
R&AN: Machine vision has become one of the defining technologies in modern robotics. What advances are you seeing in embedded vision, and how are improvements in edge processing changing what robots can perceive and respond to in real time?
KW: Machine vision has advanced greatly over the past few years, with FPGA-enabled, edge-based processing helping to filter and preprocess camera sensor data before it reaches central processors.
Offloading these kinds of depth processing, sensor fusion, object tracking, and region-of-interest detection workloads from centralized systems reduces latency and power costs, in turn making autonomous robots more accessible and reliable.
To better understand the impact of these advances, take a look at this recent collaboration between Lattice and AIRY3D. By combining a compact camera module with a dynamic FPGA, developers were able to make single-sensor 3D vision at the edge a reality.
This single-source option helps further address depth-perception and self-occlusion challenges that hinder real-world interaction, taking a meaningful step forward for execution.
R&AN: Robotics manufacturers increasingly want platforms that can evolve after deployment rather than becoming obsolete after a few years. How important are programmable hardware architectures in extending the operational life of industrial and service robots?
KW: Programmability is the key to today’s successful robotic designs. Industrial end users, for example, are investing huge sums in autonomous robotic deployments and expect equipment to last well into the future.
Why would they consider solutions with fixed capabilities when we know that, despite the progress over the last five years, there’s still more advancements to come?
Robotics manufacturers understand this, and they’re increasingly conscious of programmability when making decisions about their own facilities.
It’s up to those of us who manufacture the components that drive the robotic systems to provide options that support both pre- and post-deployment programmability, enabling lasting deployments that don’t need to be replaced the second a hardware or software model becomes outdated.
R&AN: Much of the discussion around humanoids focuses on AI models, but reliable robotics also depends on sensing, control electronics and deterministic real-time performance. Which areas of robotics hardware do you think are currently being underestimated by the industry?
KW: AI models are exciting because they are easily demonstrable; there’s a “wow” factor to any successful deployment. But what we tend to see demonstrated is, in many ways, the finished product.
It’s the result of model training and optimization at the foundation, not an out-of-the-box deployment in action. This kind of optimization needs to be a primary focus if we want to keep pushing humanoid AI models to the next level.
This isn’t the traditional “AI needs data to deliver” idea. That’s widely understood and discussed. It’s one layer deeper: AI needs the right data in the right place at the right time and showing the right things.
That starts with improvements to the infrastructure, with better power sequencing, protocol bridging, and real-time motion control. If we can build dynamic, capable, and reprogrammable components into designs at the foundational level, their downstream impact on successful AI deployment will be significant.
R&AN: Security is becoming a greater consideration as robots become connected to enterprise networks and critical infrastructure. Beyond preventing cyberattacks, what practical steps should manufacturers take to build trust and satisfy emerging regulatory requirements without adding unnecessary complexity or cost?
KW: This ties back to everything we discussed above. Security in the end product starts with secure foundational components, and the same is true for compliance. In practice, that means repositioning security within the system design. Developers must treat it as a hardware layer rather than a software-driven addition at the end of the design process.
FPGAs can provide this strong foundation, as they can execute sequencing and control logic the instant power is applied, which helps protect distributed devices from attacks that capitalize on the gap before an OS boots. They can also act as a hardware root of trust (HRoT), kicking off a trust chain as first on, last off components.
Critically, specialized options like Lattice’s MachXO3D can handle these kinds of security features in addition to other types of workloads. Bringing these capabilities into a single component helps designers improve security without complicating systems or increasing costs.
The benefits of an FPGA-based design extend to safety as well, supporting a robotic system that operates as intended without creating excess risk.
This is just as crucial for modern robotics as effective security – if a robot cannot be trusted to boot and operate reliably through various workloads, there’s no guarantee that it won’t malfunction on the job. And as these robots are deployed alongside more human workers, this kind of unexpected error is likely to pose real, physical danger.
R&AN: Looking ahead five to 10 years, which robotics sectors do you expect to create the biggest commercial opportunities for semiconductor companies? Will the growth come primarily from humanoids, industrial automation, logistics robots, autonomous vehicles, healthcare, or from markets that receive less attention today?
KW: Humanoid robotics will certainly hold significant potential for semiconductor companies; Goldman Sachs even revised its global market projection for 2035 from $6 billion to $38 billion. That’s a sign of undeniable momentum and opportunity.
But there’s certainly an opportunity for significant growth across the board. Humanoids are far from the only sector anticipating rapid growth. Analysts expect the global industrial automation market to grow to $623.25 billion by 2035 (9.13 percent compound annual growth rate) and the surgical robotics market to exceed $27 billion by 2030.
Meanwhile, certain regional AMR markets are expected to triple in size by 2030. Semiconductors are the foundation of progress across these sectors, so it’d be unwise to count them out amid the current humanoid boom.
