Distributed Cultured Neural Interfaces: External Neural Real Estate for BCI Augmentation

Building tomorrow's brain-computer interfaces through software-orchestrated biological networks

The Vision: Beyond Silicon, Beyond Single Specimens

Brain-computer interfaces have reached an inflection point. We can record from damaged neural circuits and stimulate remaining tissue, but we cannot replace lost computational capacity. When stroke, trauma, or neurodegenerative disease destroys neural real estate, current BCIs are limited to working with what remains.

Our approach is fundamentally different: instead of trying to make cultured neural tissue compete with GPUs for general computing, we're creating external neural real estate that interfaces directly with human brain circuits. Think of it as biological cloud computing for the brain - distributed, robust, and specifically designed to extend neural function for people with neurological injuries.

The key insight is architectural: rather than betting everything on single specimens (which inevitably fail), we're building networks of hundreds to thousands of cultured neural specimens that function as a collective through software orchestration. Each organoid grows in its own culture well, but they communicate through precisely controlled stimulation and recording, creating emergent computational behaviors impossible in isolated cultures.

Technical Approach

Virtual White Matter Platform: The Foundation

Our distributed architecture builds on our published Virtual White Matter platform, which enables real-time communication between spatially separated neural cultures. Unlike physical connections that are fragile and fixed, our software-defined approach allows dynamic reconfiguration of connections between specimens.

Key capabilities:

• Real-time bidirectional communication between culture wells

• Configurable delay and filtering to mimic natural neural pathways

• Scalable to arbitrary numbers of specimens

• Robust to individual specimen failures

The platform has been validated with multiple specimen types including dissociated neuron cultures, cerebral organoids, and organotypic brain slices from both rodent and human tissue.

Multi-Specimen Synchronization: Software-Orchestrated Biology

Traditional organoid computing attempts rely on single specimens, making them vulnerable to culture failures and limiting computational complexity. Our distributed approach treats individual cultures as nodes in a larger biological network, orchestrated entirely through software.

Current technical stack:

• Recording: 64-channel multi-electrode arrays (MEA; MED64, Alpha MED Scientific) with 50×50 μm² electrodes spaced 150 μm apart, utilizing four 32-channel Intan M4032 RHS stim/record headstages for simultaneous recording and stimulation

• Specimen types: (1) Embryonic day 18 (E18) rat cortical neurons cultured on PDL/laminin-coated MEAs, maintained in Neurobasal-A medium with B-27 and GlutaMAX supplements; (2) organotypic slices of adult rodent brain (thalamocortical system or hippocampus) or of adult human brain from donated neurosurgical specimens (temporal lobe cortex); (3) cerebral organoids derived from human donors; (4) aggregated neurons from a variety of sources

• Sampling and processing: Up to 30 kSamples/s recording capability with real-time signal processing on Windows 11 PC (Intel i9-12900K, 64GB RAM) running custom Python 3.0 VWM software

• Spike detection: Custom multi-threshold window discriminator system with band-pass IIR filtering (250-5000 Hz, Butterworth, 3rd order) and artifact rejection algorithms validated against MK-801 channel blocker controls

• Stimulation delivery: Current-controlled symmetrical biphasic pulses (10 μA amplitude, 500 μs pulse width per phase) delivered via Intan RHS controller with STM32F407 microcontroller for precise timing synchronization

• Cross-dish connectivity: 200 ms fixed delay between spike detection in source dish and stimulation in target dish, achieved through TCP socket communication and circular data buffering to minimize latency variability

• Validation protocols: MK-801 NMDA receptor antagonist and TTX testing to confirm biological vs. artifactual signal origin, with reversible neural activity suppression and recovery over 72-hour periods

• Machine learning integration: Real-time classification of stimulation targets using Random Forest, SVM, and K-Nearest Neighbor algorithms with 95% accuracy using 10 ms post-stimulus windows and PCA dimensionality reduction

System architecture:

• Hardware integration: 128-channel capacity utilizing bidirectional Intan RHS system with I/O expander for digital triggering, custom adapters bridging MEA mounting hardware to headstages, and Faraday cage shielding

• Software framework: Multithreaded architecture running parallel Intan RHX acquisition software and custom VWM processing with local TCP communication, circular buffering for consistent data chunk processing (25.6 ms), and real-time AI module for stimulation criteria evaluation

• Timing precision: Achieved through hardware-level synchronization via Serial Peripheral Interface (SPI) cables and microcontroller firmware, avoiding operating system variability for consistent stimulation timing

Architecture advantages:

• Fault tolerance: Loss of individual specimens doesn't crash the network

• Scalability: Add specimens without redesigning the entire system

• Flexibility: Reconfigure connections in software, not hardware

• Emergence: Complex behaviors arise from simple interactions between specimens

Current Hardware Integration

We've designed our system to leverage existing high-performance computing infrastructure while adding biological processing capabilities:

Compute tier: Intan and custom firmware handle real-time signal processing, pattern recognition, and closed-loop control algorithms

Interface tier: Custom MEA (microelectrode array) rigs with programmable stimulation, designed for 24/7 operation in sterile environments

Biological tier: Standardized culture protocols for organoids, organotypic slices, and dissociated cultures, optimized for longevity and electrical activity

Software tier: Real-time control system built on modern frameworks, designed for cloud deployment and remote operation

Milestone Gallery

Proof-of-Concept: Linking Dissociated Cultures

We've successfully linked three dissociated neuron cultures, demonstrating that stimulation in one culture can reliably trigger responses in connected cultures with controllable delays. This foundational capability scales directly to hundreds of specimens.

For illustration, we show the example of just two culture wells with electrodes, connected by software visualization enabling real-time activity propagation.

Human Cerebral Organoid Recordings

Our electrode arrays capture high-quality signals from human cerebral organoids, with signal-to-noise ratios sufficient for closed-loop control applications. These recordings demonstrate the electrical maturity needed for computational applications.

Extracellular recordings of neural activity via a 3D multi-electrode array of a human cerebral organoid (cultured 180 days in vitro). Organoids graciously donated from the laboratory of Davide Trotti, PhD, derived from specimens donated by human participants with informed consent.

Closed-Loop Human-to-Tissue Demonstration

Our most significant technical milestone: on August 23, 2025, we performed a real-time closed-loop demonstration where neural activity recorded from a human patient with intracranial electrodes- provided by Precision Neuroscience - triggered stimulation of cultured neural tissue within 100 milliseconds. This proves the fundamental feasibility of human-tissue computational interfaces.

Gaming Platform Interface

We are building a gaming platform to provide an intuitive interface for humans to interact with neural tissue while systematically exploring optimal communication strategies.

The Path Forward

Immediate Technical Challenges (6-12 months)

Scaling to 50+ specimens: We're redesigning our electrode arrays to support higher-density specimen arrays while maintaining signal quality and sterility.

Long-term culture stability: Optimizing culture conditions for months-long experiments rather than weeks, critical for practical applications.

Real-time processing optimization: Reducing I/O and data transfer latencies to achieve <10ms end-to-end response time across 1000+ channels, with GPU acceleration for parallel signal processing tasks where beneficial

Gaming platform expansion: Adding multiplayer capabilities where teams of humans can collaboratively train larger specimen networks.

Medium-term Infrastructure Needs (1-2 years)

Cloud deployment architecture: Distributing biological computing across multiple physical locations for redundancy and scale.

Standardized hardware platforms: Working with MEA manufacturers to develop standardized rigs optimized for long-term distributed experiments.

Automated culture maintenance: Robotics integration for feeding, monitoring, and maintaining hundreds of specimens.

Clinical-grade data security: HIPAA-compliant infrastructure for eventual patient applications.

Long-term Clinical Translation (3-5 years)

BCI integration protocols: Standardized interfaces between implanted BCIs and distributed cultured tissue networks.

Personalized tissue culture: Growing patient-specific organoids for optimal biocompatibility.

Regulatory pathway: Working with FDA on approval frameworks for biological computing medical devices.

Accessibility infrastructure: Ensuring the technology can serve underserved patient populations, not just those who can afford premium BCIs.

Why This Needs Infrastructure Partners

The Synchronization Challenge

Coordinating real-time stimulation and recording across 1000+ specimens requires microsecond-level synchronization across distributed hardware.  This is fundamentally a high-performance computing and I/O optimization problem that benefits from:

• Low-latency data transfer protocols and optimized I/O pipelines to minimize communication bottlenecks

• GPU acceleration for parallel signal processing computations (feature extraction, filtering)

  
  

The Data Pipeline Challenge

Each specimen generates continuous multi-channel electrophysiology data (64+ channels at 30kHz sampling). At scale, this creates massive data flow requirements:

• 1000 specimens × 64 channels × 30kHz = 1.92 billion samples/second

• Raw data rate: ~15 GB/second continuous

• Real-time processing requirements: feature extraction, pattern recognition, closed-loop control

• Storage challenges: months of continuous recording for longitudinal studies

• I/O optimization: Minimizing data transfer latencies between acquisition hardware, processing pipelines, and storage systems to maintain real-time performance

The Machine Learning Challenge

The gaming platform generates unique training datasets for biological neural networks, but analyzing this data requires specialized infrastructure:

• Temporal pattern recognition in high-dimensional biological signals

• Reinforcement learning for optimizing stimulation strategies

• Transfer learning between different specimens and culture types

• Federated learning across distributed biological computing sites

Team & Collaboration

Our interdisciplinary team combines expertise in neuroscience, bioengineering, and software development. We've published peer-reviewed research on virtual white matter platforms and filed intellectual property on key technical innovations.

We're actively seeking collaborations with:

• Infrastructure partners for scaling real-time biological computing

• Clinical partners for BCI integration and patient applications

• Gaming industry partners for expanding human-neural interfaces

• Academic collaborators for fundamental research on distributed biological computation

Contact & Next Steps

This platform represents a new approach to augmenting human neural function through distributed biological computing. Rather than competing with silicon-based AI, we're creating computational substrates that can seamlessly interface with human neural circuits.

For more information, contact us at: mdserruya @ icloud.com .

This work is supported by the Fitzgerald Translational Neuroscience Fund and conducted in accordance with IACUC and IRB approval. We're committed to responsible development of biological computing technologies with appropriate ethical oversight and community engagement.

Join us in building the future of brain-computer interfaces.