2025 Theses Doctoral

System-Level Design in the Era of Brain-Computer Interfaces

Eichler, Guy

Brain-computer interfaces (BCIs) hold significant potential to not only transform healthcare, by enabling individuals with neurological impairments to interact with the world, but also to enhance computational models and technology, opening new avenues for human-computer interaction and advancing the capabilities of artificial intelligence (AI) and machine learning (ML). However, BCI systems must overcome several obstacles to transition from research lab experiments into real-world applications.

One major challenge is the need for modern BCI systems to be mobile, wireless, and include an implanted system-on-chip (SoC) that interfaces with the brain. This requirement introduces safety concerns and imposes physical constraints on BCI systems. A second challenge is that BCI systems must integrate an ever-increasing number of sensors to support large-scale data acquisition (DAQ), enabling a better understanding of the brain. Consequently, these systems must handle growing volumes of neural data, pushing wireless transmission data rates, power consumption, and overall feasibility of implant-based BCI systems to their limits. To address these challenges, one potential solution is to integrate computation for BCI applications in specialized hardware close to the neural data source, a common approach in I/O-bound systems. Nonetheless, this solution introduces its own set of complexities, as BCI applications often rely on computationally intensive machine-learning models that must be optimized in order to be executed in real time, while meeting the physical constraints of implant-based BCI systems.

For this reason, understanding the full structure of the BCI system, including all of its components, is key for future development in the BCI field. In addition to the implanted SoC that transmits the neural data, implant-based BCI systems also incorporate a mobile, wearable SoC that receives the transmitted data. This wearable SoC must meet the physical constraints of wearable devices, which are less stringent than those of implanted devices. As long as transmission data rates between the two SoCs are sufficient, the wearable SoC can execute BCI applications either fully or in collaboration with the implanted SoC, enabling the design of practical BCI systems capable of running meaningful applications for public use.

The field of BCI is multidisciplinary and holds transformative potential for both medicine and technology. To fully realize this potential, computer architects and engineers must understand the general system-level structure of BCI systems, as well as their constraints and components. Thus, my thesis is that to unlock the full potential of the BCI field, BCI system development must be properly defined and standardized, with a clear target BCI system, an understanding of its constraints, and a distinction between three overlapping time domains that emphasize different levels of research and development, with progress continuing concurrently in each domain.

I support this thesis by presenting a three-dimensional approach to restructuring system-level design in the BCI field. Each dimension corresponds to a time domain, representing an independent area of research and development: (1) Pre-BCI – designing implant-based BCI systems that support large-scale data acquisition and wireless communication; (2) Intra-BCI – developing methodologies to integrate real-time computation and complete BCI applications into BCI systems; and (3) Post-BCI – specializing computational data flows to interact seamlessly with the biological brain.

To guide system development in these domains, I discuss the unique constraints of modern BCI systems, relying on trends in machine learning and neural interface design. I define the self-contained BCI system as our target system, capable of large-scale DAQ, high-throughput data transmission and application-level computation. I also explore potential future trends, including the use of brain-inspired neuromorphic computing as an intermediate step in the computational flow. Most importantly, I demonstrate how development can proceed concurrently across all three dimensions to overcome the key obstacles in BCI system development and accelerate the realization of fully functional BCI systems for real-world BCI applications.