Vaire Computing: Shaping the Future of Near-Zero Energy Computing

It’s not only since the recent AI hype that the immense energy consumption of data centers and its associated carbon emissions have provoked headlines. 

As the semiconductor industry strives for miniaturization, dissipating heat from densely packed chips has become a fundamental bottleneck. To avoid overheating, microprocessors today can’t operate at full capacity as not all of their parts can be powered on simultaneously, a phenomenon known as ‘dark silicon.’

Vaire Computing was founded in 2021 by Rodolfo Rosini and Hannah Earley to build near-zero energy chips. By leveraging reversible computing paired with adiabatic, gentle operation of transistors, their goal is to slash energy consumption by orders of magnitude, achieving levels of efficiency that rival the human brain. They are backed by 7percent Ventures and several business angels.

Learn more about the future of near-zero energy computing from our interview with the CEO, Rodolfo Rosini, and the CTO, Hannah Earley:

Why Did You Start Vaire Computing?

When we looked at the future of computing and Moore’s law approaching its end, we realized that fundamentally new computing architectures would be necessary to advance computing. And that one key ingredient would be reversibility. 

All of us were doing something else before Vaire Computing. Rodolfo had worked on AI startups while Hannah did her Ph.D. in molecular computing, researching how to build computers out of molecules. That’s when she first came across reversible computing, which seemed like a promising, fundamentally different way to build molecular computers. Then, one of Hannah’s thesis examiners put us in touch, and we started chatting. Reversible computing seemed like the most interesting thing no one was working on! 

Without reversibility, our computing capabilities will hit a ceiling due to heat dissipation—the next 50 years of computing will have to be reversible. If humanity likes to build a computer the size of an entire planet and calculate the answer to the ultimate question of life—just to check it’s 42—we’ll need reversible computing. 

So, we put all the other projects aside and founded Vaire Computing to build the most energy-efficient microprocessors without breaking the laws of physics. As we’ll see, this involves reversible computing as a new paradigm and a lot of hardware engineering to switch transistors more gently. 

What is Reversible Computing?

Most events in our everyday life experience aren’t reversible—ice cream usually doesn’t suddenly freeze after it has melted. Based on such observations, the second law of thermodynamics states that the total entropy of a physical system either increases or remains constant, but it never decreases. It sets the arrow of time and makes macroscopic physical processes fundamentally irreversible. 

Yet physical processes in quantum physics are fundamentally reversible. And as transistors approach quantum mechanical length scales, wouldn’t it be natural for our processors to operate also reversibly? 

We think all computers, also quantum and photonic computers, will be reversible in the future. But our starting point today are electronic microprocessors based on MOSFET transistors to switch between zeros and ones. So, we’re fully compatible with established semiconductor manufacturing. However, what we do differently is to operate those transistors more gently and use them to implement reversible computing. 

Every time a microprocessor executes a logic gate today, it generates heat. Some amount of this heat will be generated because to flip a transistor’s state, you need to move electrons, and part of the electrical energy will be converted to heat. But there’s also a fixed amount of heat generated simply because most logic gates today are irreversible. 

Consider the AND-gate as an example: it projects two input bits onto one output bit. This means you’re losing one bit of information in the process. Every time you lose information, you increase entropy and will inevitably produce a certain amount of heat, which is the essence of the Landauer principle. Since irreversible gates lose bits, they generate a fixed amount of heat simply because they’re irreversible. With billions of logic gates executed every second, that’s a lot of heat!

We need to change which logic gates microprocessors use, as they all need to be reversible, i.e., provide a one-to-one mapping between input bits and output bits. That’s why reversible computing requires rebuilding large parts of the hardware and software stack. But that’s not the only challenge ahead. 

What’s the Challenge with Reversible Computing?

It is not just that the logic gates we’re currently using are irreversible. Most calculations today are also fundamentally irreversible. As an example, consider adding two integers, A and B. If you just want to know the result, A+B, it’s impossible to know which integers A and B were added together—there are infinitely many combinations of integers whose sum is A+B. 

To be reversible, you need to calculate and store some extra information. In our example, the result of reversible addition would be A+B and A—storing the input A in addition. With this additional information, you can find out what the original A and B were, but the price is that you have to store A, which you’re not interested in, and that’s why it’s called ‘garbage.’ 

As the calculation continues unchecked, you accumulate more and more garbage until your memory overflows. If you override memory cells and thereby delete information, you’re expending energy, so you’re not saving energy overall. That’s why Rolf Landauer dismissed reversible computing immediately as impractical when he researched heat generation in computing. But there are ways to get around this problem. 

How Do You Make Reversible Computing a Reality?

The trick to avoiding the memory overflow problem in reversible computing is to do every calculation twice: First, you perform the calculation forward, storing all the bits and never wiping out information. Once you arrive at the final answer, you make a copy of it. Then, you run the entire calculation backward to get to where you started and eliminate all the garbage. This also means you only need to store one step backward. 

You might think, wait a minute, that’s actually twice the computational effort. Is this really worth it just to save some energy? 

First, doing the calculation forward and backward is the simplest approach to dealing with garbage. Over time, researchers discovered more sophisticated ways to reuse garbage bits in future computations, thereby reducing their number by orders of magnitude. 

Second, it’s hard to underestimate how much of a problem heat is in computing. Making transistors even smaller is not only extremely hard but won’t help much if we can’t use all of them for computing because the chip would overheat. 

When transistors became smaller, their energy consumption decreased proportionally so that the overall power density of a microchip stayed constant—this is called Dennard scaling. Yet, around 2005, Dennard scaling broke down. While transistors still got smaller and increasingly more were placed on microchips, the clock speed of microprocessors plateaued around a few GHz instead of reaching dozens or hundreds of GHz.

Image Credits: Karl Rupp, CC BY 4.0 (Source)

In the worst case, reversible computing requires twice as much circuitry, but it will allow us to build much larger and, therefore, much more powerful computers. The key to getting there is to avoid heat dissipation as much as possible. 

Besides reversibility, the other key ingredient is reducing the heat generated by switching transistors. Transistors are normally switched as fast as possible, typically within tens of picoseconds, as electronic signals take the shape of square waves. We use the very same MOSFET transistors but switch them more gently so that the process becomes adiabatic, i.e., avoids heat exchange with the environment. This engineering challenge is at the core of what we do on the hardware side. 

On the software side, we’ll need to make reversible computing compatible with the software and software development process we have today. If you learned how to program and switched between different programming paradigms, e.g., from imperative to functional or object-oriented programming, you know it took some mental effort to get used to it. We want to avoid people having to learn reversible programming, and instead, we need to deal with it when we compile software for our chip and how our chip operates.

If a computer could be made entirely of reversible gates and operate completely adiabatically, it would not dissipate any heat from computation. In practice, some energy is always lost due to non-ideal characteristics of real materials and devices. But adiabatic switching of transistors combined with reversible computing will allow us to reduce energy consumption by several orders of magnitudes—more energy efficient than the human brain. That’s why we call our chips Near-Zero Energy Chips (NZECs).

Image Credits: Mike Frank, Sandia National Laboratories

How Did You Evaluate Your Startup Idea?

We trade single-core for multicore performance—operating transistors adiabatically impacts single-core performance, but being super energy-efficient will allow us to stack compute layers on top of each other and build truly 3D chips. Microprocessors today are 2D because if you stack them, they will trap even more heat inside that you can’t get rid of fast enough, and they will overheat. 

Building 3D chips with many compute layers will be a game-changer, so naturally, our starting point will be GPUs and machine learning as the first use case. Being super energy efficient will allow us to address various edge applications. 

We expect that improving energy consumption by several orders of magnitude will unlock a market potential bigger than the entire chip industry. The number of computers has steadily increased, from the first computer in the world to one computer per company to several computers per person. Next, we’ll have hundreds of computers per person. We’ll have machine intelligence everywhere and always-on AI devices that won’t depend on batteries but can use capacitors. 

What Advice Would You Give Fellow Deep Tech Founders?

Don’t do a chip startup. There’s a reason venture capitalists have moved into software, and no one has ever returned. Just kidding. 

It’s actually the opposite. Using software has been a key differentiator for companies, but with the cost of cloning software dropping to zero thanks to AI, it is now becoming a commodity. In the same way that every company uses electricity or the internet, using software and AI will be taken for granted. Instead, domain knowledge and solving hard problems in hardware will become valuable—deep tech is on the rise! 

Finally, you only have a certain number of companies that you can start in your life, so choose wisely. Over time, you’ll realize that it doesn’t make sense to work on something that is just profitable but boring and has no impact. Instead, seek the hardest and most impactful problem you could be working on. It’s not ‘just follow your passion’—you have to find the intersection of what you love and what matters. You won’t cure cancer or build space stations if you work for McKinsey.

Want to Shape the Future of Near-Zero Energy?

Vaire is looking for great people to join their team and achieve what the industry deems impossible. Take a look at their Notion page, Working at Vaire Computing*, and consider getting in touch with your CV and cover letter.


*Sponsored link—we greatly appreciate the support by Vaire Computing

Comments are closed.