Probably everyone who has ever used a computer probably knows that today’s consumer computers are far from perfect. Unlike our brains, they are not able to learn the most efficient processes and must repeat them at the same or lower speed over time. Moreover, they can send information from one place to another at one time. However, it could change the so-called neuromorphic chips, or neuromorphic computers made up of these chips.
In the spirit of the brain, neuromorphic chips are able to “learn” through architecture analogous simulating elements of the brain (neurons, synapses), which processes need to be performed more often and which less often, or more, can perform multiple operations at once.
SpiNNaker (Spiking Neural Network Architecture) is the latest iron in the development of similar computers, capable of executing over 200 trillion parallel operations – a figure of 12 zeros. However, even this set is still insufficient compared to what each of us wears in the skull. SpiNNaker, with all its capacity, reaches “only” a thousandth of a percent of the estimated speed of the human brain.
However, its authors from the united British-European team intend to gradually reduce the lead of homo sapiens – the next goal will be the development of SpiNNaker’s successor reaching perhaps a thousand times higher speed. Thus, such a machine could have one percent of the speed of the human brain, at least as it is possible to compare difficult to compare systems side by side.
What is the use of such operations? In addition to advances in the field of computer science, especially for simulations, respectively. rather, emulation of biological processes. In the summer of this year, SpiNNaker made its first emulation of tens of thousands of neurons linked into a single unit. His future simulations are expected to better understand the functioning of the human brain – and perhaps to better understand neurodegenerative diseases such as Alzheimer’s and Parkinson’s. In this case, the development of informatics goes hand in hand with the possibilities for medicine.
This does not mean that neuromorphic computers will only be useful for doctors. The advantage of parallel operation is the possibility to realize calculations several times more economically in terms of energy output. In other words, neuromorphic chips have the chance to reduce the current power demand of computers and thus their cooling.
Tomorrow’s neuromorphic chips could bring machine learning applications to mobile computers. It is a bit exaggerated to say that, similarly to PC owners, in the past, they purchased designed graphics cards, in the future, we can look forward to specialized “AI cards”.
Příslib dramaticky narůstající výpočetní kapacity příštích dekád je o to větší, že spolu s neuromorfními počítači pokračuje kupředu i vývoj kvantových počítačů, počítačů využívajících uhlíkových nanotrubic a dalších technologií. Ty by mohly zajistit, že by informatika po roce 2030 mohla dalece přesahovat dnešní možnosti. Paradoxem je, že SpiNNaker jako takový je v tomto ohledu už vlastně mírně zastaralý – jeho konstrukce trvala přes 10 let a započala již v roce 2006. Vývoj technologie pak začal v polovině 90. let.
Dnes vyvíjené neuromorfní čipy jako Spikey by jej v součtu mohly značně překonat. Než je však někdo smontuje do použitelných superpočítačů, může to ještě nějaký čas trvat. Ostatně ani SpiNNaker nepostavili vědci za den.