A new computing paradigm- thermodynamic computing- has entered the scene. Ok, okay, maybe it’s just Potential computation With a new name. They use both noise (such as due to thermal ups and downs), instead of fighting it, to calculate. But still, this is a new physical approach.
“If you are talking about computing paradigms, no, it is the same computing paradigm,” as possible computing, says Beatsh Behin-OyeFounder of CTOs and Possible Computing Startup Ludwig computing (In the name of Ludwig Boltzman, a scientist is largely responsible for the field, you guess, thermodynamics). “But this is a new implementation,” he says.
recently Publication In Nature communicationNew York -based Startup General computing Before what they call thermodynamic computer, expand the prototype before. They have demonstrated that they can use it to exploit the noise to reverse the matris. He also demonstrated the Gausian sample, which underlines some AI applications.
Noise can help some computing problems
Traditionally, the noise is the enemy of calculation. However, some applications really rely on the noise generated. And using naturally occurring noise can be very efficient.
“We are focusing on algorithms that are capable of taking advantage of noise, stochastic and non-deteriorism,” says Zachery beltakeSilicon engineering lead in general computing. “This algorithm space becomes very large, from scientific computing to linear algebra. But a thermodynamic computer is not helping you to check your email soon.”
For these applications, a thermodynamic-or potential-computer begins with its components in some semi-disciplinary state. Then, the program is programmed in interaction between the problem components trying to solve the user. Over time, these interactions allow components to come into balance. This balance is a solution to calculation.
This approach is a natural fit for some scientific computing applications that already include randomism, such as Monte-Karlo simulation. This AI image generation is also well suited for algorithm stable proliferation, and a type of AI known as potential AI. Surprisingly, it appears to be well suited for some linear algebra computations that are not naturally probable. This applies the approach more widely to AI training.
“Now we see with AI that the paradigm of CPU and GPU is being used, but it is being used because it was there. Nothing else was there. It is said that I have found a gold mine. I basically want to dig it. Do I have a shovel? Or do I have a bulldozer? Mohammad c. BojachaluiCEO and co-founder of Ludwig Computing. “We are saying that this is a separate world that requires a separate equipment.”
General computing approach
The prototype chip of general computing, called the stochastic processing unit (SPU), consists of eight capacitor-Indian resonators and random noise generators. Each resonant is connected to each other through a tune -tuned coupler. The resonance is initiated with randomly generated noise, and under the study the problem is programmed in coupling. After the system reaches the balance, resonant units are read to get a solution.
“In a traditional chip, everything is very controlled,” says Gavin crookAn employee research scientist in general computing. “Take your leg slightly away from control, and the matter will naturally start behaving more stochastic.”
Although it was a successful proof-off-concept, the general computing team admits that it is not prototype scalable. But they have Amendment Their design is getting rid of tricy-to-scale inductors. They are now planning to create their next design in Silico instead of a printed circuit board, and expect their next chip to come out at the end of this year.
How far can this technique be seen. The design is CMOS-compatible, but a lot of work has to be done before being used to solve real-world problems. “It’s surprising what he has done,” says Ludwig Computing’s Bozlui. “But at the same time, a lot of work can be done to actually take it that is something for the commercial product from today that can be used on scale.”
A different vision
Although potential computing and thermodynamic computing are essentially the same paradigm, a cultural difference. Companies and researchers working on potential computing almost specifically detect their educational roots Supriyo dutta At the University of Purdue. Three cofounders of general computing, however, have no relations with perdue and come from the background in quantum computing.
This is a slightly different vision in normal computing cofounders. They imagine a world where different types of physics are used for their own computing hardware, and every problem needs to be solved, coincides with the most optimal hardware implementation.
“We coined the word physics-based asics,” referring to Beltch of general computing, ” Application-specific integrated circuitIn his view, a future computer will have access to traditional CPU and GPU, but a quantum computing chip, a thermodynamic computing chip, and any other paradigm people can dream. And each calculation will be sent to an ASIC that uses physics which is best suited for the problem in hand.
From your site articles
Related articles around web
