It occurs to me that one might utilize hemispherical atomic clusters bound on their flat sides to some substrate to mimic the behavior of elements of the neural network. In my work on the mathematical motivation in atomic shell structures, I found that in simple harmonic oscillator (basically 3D music) models of spherical atomic nuclei, I discovered that each successive shell had a set of simple mathematical relationships. The total numbers of nucleons in each shell was always a successive doubled triangular number, so 2, 6, 12, 20, 30, 42.... And it's been known for many decades that the energy levels of each shell always increased by one, so that the 1s level, with 2 nucleons, had the lowest energy of 1.5 h-bar omega units, then 1p has 2.5, 1d2s has 3.5, and so on. I then discovered that the TOTAL shell energies (multiplying the energy levels by their total nucleon occupancy) gave values whose differences (between successive shells) was always 3x a square number, so 3x(1, 4, 9, 16)... etc. I would imagine one could think of this as analogous to the multiplication of the WEIGHTS of neural-network connections...etc. Now atomic clusters have many properties in common with atomic nuclei, including shell structures in the harmonic oscillator model. And by altering their constituencies in terms of atoms of different elements, one could, in theory, imitate the electronic properties of any pure element. In the end, one might have an entire system of atomic clusters forming your neural network, if one can work out how to interconnect them. Optically might be one way- by varying the frequencies of light emitted, and adjusting the transparency of the clusters so they can absorb/emit now one frequency, now another, and in different places in the cluster at different times. Imagine how tightly packed a system like this could be.....
There is an article in the latest (29 July 2022) Science on analog-switching "neurons" that are 1/1000th the size of human neurons.
ReplyDelete--Frank
Thanks, Frank. I know people have been working on analog and 'neuromorphic' neurons for some time.
DeleteIt occurs to me that one might utilize hemispherical atomic clusters bound on their flat sides to some substrate to mimic the behavior of elements of the neural network. In my work on the mathematical motivation in atomic shell structures, I found that in simple harmonic oscillator (basically 3D music) models of spherical atomic nuclei, I discovered that each successive shell had a set of simple mathematical relationships. The total numbers of nucleons in each shell was always a successive doubled triangular number, so 2, 6, 12, 20, 30, 42.... And it's been known for many decades that the energy levels of each shell always increased by one, so that the 1s level, with 2 nucleons, had the lowest energy of 1.5 h-bar omega units, then 1p has 2.5, 1d2s has 3.5, and so on. I then discovered that the TOTAL shell energies (multiplying the energy levels by their total nucleon occupancy) gave values whose differences (between successive shells) was always 3x a square number, so 3x(1, 4, 9, 16)... etc. I would imagine one could think of this as analogous to the multiplication of the WEIGHTS of neural-network connections...etc. Now atomic clusters have many properties in common with atomic nuclei, including shell structures in the harmonic oscillator model. And by altering their constituencies in terms of atoms of different elements, one could, in theory, imitate the electronic properties of any pure element. In the end, one might have an entire system of atomic clusters forming your neural network, if one can work out how to interconnect them. Optically might be one way- by varying the frequencies of light emitted, and adjusting the transparency of the clusters so they can absorb/emit now one frequency, now another, and in different places in the cluster at different times. Imagine how tightly packed a system like this could be.....
ReplyDelete