# Is it possible to construct 3-dimensional microchips?

In our daily computer, the microchips in it are 2-dimensional, but could it be possible to produce 3-dimensional ones, in some way similar to the brain structure?

• I was thinking about this and previous question. I didn't wanted to bore people with details, and making pure psychological answer on CS site, but very briefly (at least for now): you cannot manufacture brain like structure and comparing NN to brain fails on several levels. On the both ends (Psychology and Computer Science) there are issues that need to be resolved, because for now this is impossible. Neural Networks even those fancy ones with biological roots and "brain mimic" structure are not even oversimplified model of the brain. Moreover any nonadaptive construction will fail. – Evil May 4 '16 at 0:38
• We already have 3D microchips. (e.g. 3dnand for ssd). Creating 3d logic chips is hard because of the thermal heat produced. A 2D CPU already produces 100W+; also the added complexity in 3D logic vs 2D logic would render the exercise uneconomical. For 3D RAM it's different. The structure is so simple and repetitive it works. – Johan - reinstate Monica May 4 '16 at 14:29
• The heat production was also one of the issues I was thinking about. But that´s because the whole circuit has just one supply of energy (or some more), while in the brain, each of the cells has it´s own energy supply, so heat is evenly distributed over the structure, in contrast to the microchip. Is it possible to physically make neuron-kind structures and their connections, instead of programming the workings of a neuron, and programming the connections between them? – descheleschilder May 4 '16 at 18:21
• Maybe you care about 3D IC on Wikipedia. – Juho May 5 '16 at 8:40

Yes, it's possible to build 3-dimensional integrated circuits: we already know how to do that, and many chips already use 3D integration.

Building chips whose structure is similar to the brain structure is a different question. Circuits on current chips (whether 2-dimensional or 3-dimensional) are very different from the structure of the brain.

I don't see the point. For better neutral networks, you need better connectivity between processors, which can be done with parallel processing. You also need machine learning algorithms that are o optimized for the hardware. But, you could theoretically build a processor, many of them inextricably linked to each other for better processing. Really, only time will tell.

• Of course, the architecture of a 3-dimensional chip is very different from that of the brain. But would it nonetheless be possible to construct chips that dó have the same structure: artificial cells that have their own energy source, and connect each cell with thousands of others while the interconnectivity between the artificial cells can vary in strength, just like the connections between different neurons. Or can this kind of architecture (where the connections are very similar to lightning, or tree branches and -roots?) only develop in a Natural way in respons to a very chaotic world? – descheleschilder May 4 '16 at 17:46

The latest chips are from more than 10 layers. The layers are constructed essentially with advanced photography.

In the earlier eras of the chip manufacturing, there was also a different technology common: ion implantation. It works by shooting the target wafer with accelerated ions. Their penetration depth was much higher as the current photographyical technologies. The technology has became rare today, mainly on engineering and cost/profit predictability reasons.

However, the current chip technology has a limit: to make photography, we need to be able to mirror photons. With ultraviolet light it is yet possible, but with rontgen, it is practically not. There is no matter what could reflect it, and it has unavoidable physical reason. The current last developments are going into the high ultraviolet spectrum, with extreme costs. After that, the development will stop.

However, also the number of the layers of the current chip production could be increased, or the ion implantation could get space again. Thus, a major development into this direction is not closed out in the future, although the current cutting edge is the tuning of the EULV.

You would have problems with their cooling - a $\mathrm{cm}^3$ silicon cube, from millions of layers, can be cooled like a $\approx \mathrm{cm}^2$-sized traditional chip. Having a such a "3d-chip", it should be cooled by some very exotic way (like having a lattice of cooling channels in it), or the chip itself should be very exotic to avoid the heat problem.