Neural Computer Architecture: Bridging Biology and Technology

As part of my ongoing research collaboration between the University of Zagreb and Beijing Normal University, I’m working on implementing basic cognitive mechanisms in technically feasible systems. This project represents a fascinating intersection of neuroscience and computer engineering.

The Biological Inspiration

Traditional computer architectures follow the von Neumann model, with clear separation between processing units and memory. However, biological neural networks operate fundamentally differently:

• Distributed processing: Each neuron simultaneously acts as both processor and memory
• Synaptic plasticity: Connections strengthen or weaken based on activity
• Content-addressable memory: Information retrieval based on content patterns rather than addresses

Technical Implementation

Our approach focuses on creating biologically-restrained neural networks that can perform computational tasks while maintaining biological plausibility. Key features include:

Synaptic Plasticity Models

We implement Hebbian learning rules and spike-timing dependent plasticity (STDP) to enable adaptive behavior without external supervision.

Content-Addressable Memory

Using associative memory principles similar to Hopfield networks, our system can retrieve stored patterns based on partial or noisy input cues.

Current Challenges

The main challenges we’re addressing include:

1. Scalability: How to maintain biological constraints while scaling to useful computational sizes
2. Learning efficiency: Balancing biological realism with practical learning speeds
3. Hardware implementation: Exploring FPGA-based implementations for real-time processing

This research has direct applications in neuromorphic computing and could lead to more efficient, brain-inspired computational systems.

This work builds on my paper “Biological Neural Machines – Creating a Computer on Biologically Restrained Neural Networks” presented at the IRCN and Chen Institute Joint Course on Neuro-inspired Computation, 2025.