As we advance further in building AI models it seems that the Von Neuman architecture has some certain limitations. In a real-life scenario, neurons work in bulk and information is stored in networks. Neurons have thousands of input and output connections with other neurons with some of them being weak while others are strong. When they fire together, a signal based on the weights of the connection path is created and that causes a pattern of other neurons firing back in response. There are no single units that store information.
The major distinction is that there is not discrepancy between storing/retrieving information and computation as in Von Neuman model.
- Is there any system currently in the market or research sectors that uses a distinct architecture?
- Can anyone refer or propose briefly, in a simple manner, a different framework?
As a starter, I would recommend a very interesting article that talks about this in a broad view: https://medium.com/swlh/the-explosion-of-new-architectures-is-fundamentally-changing-computing-f69b7faae89d
On a more technical level, I can recommend the work of Ganguly et al. https://ieeexplore.ieee.org/abstract/document/8697354/
I hope these are of interest to you.