Ask HN: Is neuromorphic computing going to replace traditional AI?

5 points by lennart-rth 58 days ago | 3 comments

I’ve recently learned about neuromorphic computing while thinking about the core inefficiencies of current deep learning.

Today AI relies heavily on super-dense layers, continuous computation of all layers and neurons at all times, and global backpropagation to find optimal weight updates. But when you look at the human brain, none of that happens.

The brain operates on principles that stand in contrast to modern LLMs:

- Local Evolution: Neurons are largely independent, evolving based on their local neighborhood and simple feedback loops like neurotransmitters (e.g., dopamine) rather than a global error signal. - Extreme Sparsity: The system is massively sparse (neurons only evolve and get updated when they have been involved in a firing-chain). - Event-Driven Processing: Neurons are only firing when they get evaluated and are actually triggered by other spiking neurons.

In contrast, current LLM‘s seem to me like they brute-force their way through problems by training all layers and neurons at the same time and trying to find the optimal global update function.

If you look at the trajectory of AI advancements over the last few years, a clear pattern emerges:

- 2020–2022: Scaling up datasets and raw compute. - 2023–2024: Expanding context windows and shifting to Mixture of Experts (MoE). - 2024–2025: Chain-of-thought and inference-time reasoning. - 2025–Present: Autonomous execution and parallel multi-agent systems.

Fundamentally, every single one of these advancements is just a different way of scaling up compute and processed tokens. So this scaling will naturally hit a wall at some point as electricity is not unlimited. That’s why I believe that long term progress can not come from just scaling forever. What it needs is radical efficiency improvements. So could neuromorphic processing be exactly that or is it not mature yet?

mikewarot 56 days ago |

No, it won't. The problem is that the efficiencies of analog multiplications get swamped by the need to digitize the outputs at some point. Converting back to digital with many bits of precision is a slow, energy consuming process.

If you want to take a single analog voltage, and convert it to 16 bits, for example, at 80 million samples per second, the LTC2163[1,2] will do it, but it's a $120 chip, and consumes 108 mWatt to do it. Lower resolutions, and slower speeds cost less, but still that's the major problem you encounter.

[1] https://www.digikey.com/en/products/detail/analog-devices-in...

[2] https://www.analog.com/media/en/technical-documentation/data...

plastic-enjoyer 57 days ago |

> So could neuromorphic processing be exactly that or is it not mature yet?

So, take this with a grain of salt. I only dabbled in Computational Neuroscience and met some people working on neuromorphic computing, but I am not an expert.

So, the short answer is, that it is not mature yet, and there is currently no indication that this will change in the foreseeable future. Neuromorphic chips are built for spiking neural networks, not the matrix multiplications behind LLMs and CNNs. The promise are huge improvements in energy efficiency for edge stuff like sensors and robotics, but SNNs still can't match standard deep learning on the benchmarks that matter. The software ecosystem is nowhere near PyTorch/CUDA, there was no AlexNet moment for SNNs, hardware access is limited to a few research groups, and funding follows from that. So, as you can see, it is a bit circular. If you are interested in neuromorphic computing, have a look at SpiNNaker2

Gooblebrai 57 days ago | |

Thanks for the comment. Even if they don't compete with traditional LLMs and CNNs, do the people you spoke with see any potential breakthrough application of this type of chip?