Opportunities and challenges of brain-on-a-chip interfaces
image: The various types of BoCI are illustrated. The various types of BoCIs are introduced, including planar and stereoelectrode-based methods. The former primarily relies on low-density or high-density planar microelectrode arrays (MEAs), while the latter employs techniques such as 3-dimensional (3D) MEAs, implantable probes, and symbiotic electrodes to capture signals from different layers and regions of brain organoids. LD-MEAs, low-density MEAs; HD-MEAs, high-density MEAs.
Credit: Dong Ming, Academy of Medical Engineering and Translational Medicine, Tianjin University.
BoCIs differ fundamentally from traditional brain-computer interfaces (BCIs) by using in vitro “lab-grown brains”—including 2D neural networks, 3D brain organoids, and brain slices—instead of living human or animal brains. This design offers unparalleled experimental controllability, scalability, and ethical compliance, making it a powerful platform for studying neural function and developing bio-inspired computing systems.
At the heart of BoCIs is the integration of biological neural networks with advanced electrode systems, enabling precise signal recording and stimulation. Two primary interface designs dominate current research: (1) Rely on 2D electrode grids to detect extracellular electrophysiological signals (e.g., action potentials, local field potentials) from neural networks. (2) Low-density MEAs (64–256 electrodes) support network-level analysis, while high-density MEAs (HD-MEAs) with up to 26,400 electrodes (pitch as small as 0.25 μm) enable single-cell and synapse-level resolution. (3) Ideal for long-term, non-invasive recording of 2D neural networks and brain organoids plated directly on electrode surfaces. (4) Address the limitations of planar MEAs for 3D brain organoids, which require depth-resolved signal capture. (5) Four key designs: 3D MEAs (protruding electrodes penetrating 40–100 μm into tissue), implantable probes (for deep signal recording), wrapped MEAs (flexible baskets or shells conforming to organoid shapes), and symbiotic electrodes (integrated with organoids during development for non-invasive, long-term monitoring).
BoCIs enable dynamic interaction between biological networks and external systems through tailored stimulation and training paradigms: (1) Electrical stimulation: Encoded via amplitude, frequency, or spatial patterns to promote neural maturation, synaptic plasticity, and learning. Low-frequency (0.2–1 Hz) stimulation induces memory formation, while high-frequency (10–20 Hz) pulses trigger long-term potentiation (LTP) or depression (LTD). (2) Optical stimulation: Uses optogenetics to target specific neurons, avoiding electrical artifacts and enabling precise spatiotemporal control. (3) Chemical stimulation: Modulates neural activity via neurotransmitters but is less commonly used due to lower predictability. (4) Open-loop training: Employs fixed stimulation parameters to reshape neural network topology and strengthen connections. (5) Closed-loop training: Dynamically adjusts stimuli based on real-time neural activity, enhancing learning efficiency. This approach has enabled lab-grown brains to control robots, play video games, and perform complex tasks within minutes.
BoCIs leverage the inherent advantages of biological neural networks—ultralow energy consumption, dynamic plasticity, and efficient learning—to drive innovative applications: (1) Mobile robots: Lab-grown brains have controlled robots for obstacle avoidance, maze navigation, and target tracking by decoding neural signals into movement commands. (2) Robotic arms: Closed-loop systems enable adaptive behaviors like drawing, with neural activity guiding precise motor control. (3) Virtual tasks: Neural networks have learned to play the game Pong and control simulated aircraft, demonstrating real-time decision-making and feedback integration. (4) BoCIs merge biological intelligence (e.g., learning, memory) with artificial intelligence (AI) algorithms to create more efficient computing systems. (5) As “physical reservoirs” in reservoir computing, lab-grown brains excel at pattern recognition (e.g., spoken digit classification) and nonlinear problem-solving, outperforming traditional silicon-based AI in data efficiency and energy use. (6) Integration with reinforcement learning (e.g., EXP3 algorithm) and spiking neural networks has enabled rapid task adaptation, such as training neural networks to play Pong in 5 minutes.
Despite significant progress, BoCIs face four foundational challenges: (1) Optimizing the “Intelligent Foundation”. (2) Engineering High-Fidelity Bioelectronic Interfaces. (3) Enhancing Neural Plasticity and Learning. (4) Integrating Biological and Artificial Intelligence.
BoCIs represent a paradigm shift in hybrid intelligence, merging the computational efficiency of machines with the adaptive, low-energy properties of biological neural networks. “By addressing current limitations in organoid maturity, interface technology, and AI integration, BoCIs could revolutionize biocomputing, robotic control, and our understanding of neural function,” noted Dr. Dong Ming. The review calls for interdisciplinary collaboration across engineering, neuroscience, synthetic biology, and ethics to unlock BoCIs’ full potential.
Authors of the paper include Wenwei Shao, Weiwei Meng, Jiachen Zuo, Xiaohong Li, and Dong Ming.
The study was funded by the National Key Research and Development Program of China (2021YFF1200800) and the National Natural Science Foundation of China (82101853, 82171861, and 81971782).
The paper, “Opportunities and Challenges of Brain-on-a-Chip Interfaces” was published in the journal Cyborg and Bionic Systems Jun. 17, 2025, at DOI: 10.34133/
cbsystems.0287.