Max Aifer
@MaxAifer
Followers
3K
Following
5K
Media
77
Statuses
548
Theorist @NormalComputing. Thermodynamic computing for efficient AI.
New York, NY
Joined July 2021
Here’s a thread on our new perspective article Solving the compute crisis with physics-based ASICs. To meet the demands of AI, a variety of new approaches to computing have been proposed that reimagine the basic building blocks of computers by considering the intersection of
4
10
59
I'll be on a stream with @InferenceActive on august 6 to talk about AI and chips that exploit physics. Come check it out!
3
10
30
space to discuss the paper
Here’s a thread on our new perspective article Solving the compute crisis with physics-based ASICs. To meet the demands of AI, a variety of new approaches to computing have been proposed that reimagine the basic building blocks of computers by considering the intersection of
10
4
20
Btw, I’m planning to host a space on this tonight.
Here’s a thread on our new perspective article Solving the compute crisis with physics-based ASICs. To meet the demands of AI, a variety of new approaches to computing have been proposed that reimagine the basic building blocks of computers by considering the intersection of
2
1
16
Big thanks to my awesome coauthors @blip_tm @BramhavarSuraj @KeremCamsari @ColesThermoAI @gavincrooks, Douglas J. Durian, Andrea J. Liu, @amarchenkova @zaqqwerty_ai @peterlmcmahon @FarisSbahi, Ben Weiner, @LoganGWright1 !!.
0
1
7
While there is promising early progress in the field of physics-based ASICs, much remains to be done! In particular, we highlight several calls to action in the perspective, including. - Identifying a set of applications for which GPUs are not well optimized. - Co-designing.
2
0
7
It is not immediately obvious why the overlap between these two categories should be large or even non-empty; however, it is fortunate that there appear to be a large number of applications for physics-based ASICs. For example, a number of problems in linear algebra that can be
1
2
6
As an intuition for why physics-based ASICs may be dramatically more effective in some domains than general-purpose digital architectures, consider the problem of simulating a single transistor. A faithful physical simulation of a single transistor would require billions of
1
2
7
Many computational paradigms have been proposed that can be interpreted as physics-based ASICs, such as analog computing, thermodynamic computing, Ising machines, stochastic digital computing, and optical computing. Frequently, such devices make us of stateful logical components
1
1
6
As a result, there has been increased interest in achieving performance gains by developing new architectures and paradigms, rather than just scaling transistor count, and a number of basic assumptions underlying conventional circuit design are being reconsidered. Integrated
1
1
6
While AI continues to develop rapidly, it does so with massive hardware and energy costs. Until the early 2000s, computers predictably gained in both speed and energy efficiency due to the combination of Moore’s law and Dennard scaling. In the post-Dennard world where performance
1
1
6
RT @wjzeng: Looking forward to seeing the physics-based ASIC field grow further. Great overview from @NormalComputing and collaborators.
0
2
0
RT @amarchenkova: Really excited to announce our new preprint, "Solving the compute crisis with physics-based ASICs!" For years, I’ve argue….
0
8
0
RT @NormalComputing: We’re excited to announce our preprint "Solving the compute crisis with physics-based ASICs"!. Our team at @NormalComp….
0
17
0
RT @KeremCamsari: Check out this bold new work from @NormalComputing — forecasting a golden age of computing driven by physics-based hardwa….
0
4
0
The era of Physics-based ASICs has arrived. On the surface, computing infrastructure seems unready to meet the demands of AI. But underneath, a Cambrian explosion of new computer architectures is rising to meet this challenge, with the goal of using physical processes as a
12
36
229
RT @KlettPhoebe: There will be much fanfare at our next Proof Liturgy. As always, all math curious are welcome!
0
4
0
RT @thomasahle: VerilogEval is an odd coding dataset - it gives the agent access to the golden testbench, which you will use to evaluate th….
0
7
0
RT @ColesThermoAI: Interesting paper by Stephen Whitelam at LBNL that builds on our work on thermodynamic computing. In 2023, our team at….
0
3
0