Lattice VP on FPGA security for robots and humanoids
Lattice Semiconductor's Eric Sivertson discussed using FPGAs as a root-of-trust to secure robots, especially humanoids, against cyberattacks that could
Eric Sivertson, vice president of the security business at Lattice Semiconductor, says securing robots requires a new paradigm. He argues that field-programmable gate arrays, or FPGAs, are critical for locking down systems, especially as robots like humanoids move out of cages and into shared spaces.
Sivertson made the comments on The Robot Report Podcast. The conversation focused on security risk vectors in robotics. He used humanoids as a standard-bearer for the intersection of AI and the physical world.
The safe and secure change
Sivertson stated that you cannot trust a humanoid if it is not both safe and secure. This represents a major shift. Many current robots operate in cages or are locked to the floor, limiting mobility. Humanoids, and systems like Tesla vehicles which he called "rolling robots," change the equation. These cyber-physical systems combine AI with the physical world.
Safety parameters are not enough if they can be compromised. "You've built all the safety parameters, but if someone can go in now and adjust the data set that those safety parameters are using to make decisions, it's not really safe anymore," Sivertson said. A cyberattack on a thousand humanoids could turn them into weapons moving 50- to 100-kg items. This fusion of AI with the physical world will force the integration of safety and security, which have traditionally been separate concerns.
Key security features for hardware
Sivertson outlined several key security features hardware developers must consider. Attestation is primary. Systems must verify that connected components, like a PLC or sensor, are trusted entities. The verification must be mutual. A sensor must confirm it is communicating with the manufacturer's true cloud.
The next layer involves ensuring both ends are running authenticated code using digital signatures. Sivertson compared this to a pilot verifying his identity and then running through a pre-flight checklist. Once entities and configurations are authenticated, data encryption can protect communications during operation. This makes it hard to intercept or spoof data.
Where FPGAs can help
Sivertson explained the architectural role of FPGAs, particularly Lattice's root-of-trust products. A key differentiator is non-volatile memory contained within the chip. This allows the chip to boot from two different locked flash images. Once the chip leaves the factory, its configuration can be locked down to prevent corruption. A known-good image can supervise updates to the other, making attacks difficult.
This design addresses a critical vulnerability. Main processors, AI engines, GPUs, and CPUs in advanced geometries lack internal flash memory. They must fetch instructions from external non-volatile memory or the cloud. That boot process is highly vulnerable to denial-of-service or ransomware attacks.
"You don't want denial of service when your car's driving down the road and is gonna automatically turn, and it keeps going off a cliff," Sivertson noted. Lattice FPGAs can monitor the boot process of these larger processors. If a GPU signals it cannot run due to a lack of authenticated code, the FPGA can act to recover the image and restore functionality.
He highlighted the cost-effectiveness of this approach. A sub-$10 FPGA part can guarantee the operation of a processor costing $100 or $1,000. This provides cyber resilience by ensuring critical components always function as intended, even under attack.
