Written by F. Ece Demirer. Photo credit F. Lemaitre
How photonics can help with the high energy consumption of the crypto currency mining activities
Even though it is hard to reason with Elon Musk sometimes, this time he is right on the point! We need to address the high energy consumption problem of crypto currencies. As a group of photonics enthusiasts from Photonics Society Eindhoven, we reached out to experts who are trying to solve the energy consumption problem through using integrated photonics approaches for crypto currency mining. Dr Bogdan Penkovsky, who is the co-founder of the non-profit organization PoWx, accepted our request for an interview. The discussion revolved around crypto currency mining and how hybrid photonic-electronic computation can help to do this ’cheaper’. We are excited to see that the hot topic of crypto currency mining is attracting attention towards photonics solutions. With numerous eyeballs watching the topic, we believe this is a great opportunity to increase the role of photonics in our daily life’s. Not to mention that it can be as valuable as mining gold, literally.
To bring everyone on the same page we briefly address why using crypto currencies consume energy at the first place and what does photonics have to do with that. I assume that by now we all heard about the term crypto currency. Probably, none of our readers are experts on this topic; that’s why we tried to do our best to give simple descriptions of some concepts. It is important note that in this article, we don’t dive into discussions of (i) whether we need crypto currencies and decentralized banking systems, nor (ii) if the actions of centralized banks such as printing money etc. is as bad as it sounds. We simply recognize that the crypto currency use is increasing, and institutions are starting to adopt it.
Starting with the question: “If crypto currency is a technological product, what do they deliver to consumers?” Banks deliver multiple services among them one role is to enable trustable money transactions by preventing misconduct, double spending and stealing. Crypto currencies replace this role by using a technology called block chain, which delivers the same functionality of a central body, without the central body itself. In this technology, the system assumes nobody is trustable. For example, one such block chain system requires that a transaction should be approved and recorded by the majority of the participants before it can take place. As a result, multiple computations take place in different parts of the world and if the answer is the same, the transaction is allowed. This justification process is called proof-of-work (PoW), and it is at the base of the majority of the crypto mining activities. The miners are awarded when they help a transaction to occur, thus when they compute. An alternative to PoW is proof-of-stake (PoS), which is suggested to decrease the energy consumptions associated with PoW activities. In this article we focused solely on the PoW. Among them, there is a large number of algorithms suggested to decrease the energy consumption without sacrificing the security of the whole system. So far, even the most efficient ones are consuming significant amounts of energy. When it comes to heating the planet to provide such a service, we need to ask ourselves: “Is there a cheaper way of doing the same thing?”’ And ‘cheaper’ in this case refers to ways that will not cost us our planet.
Let us dive into the root of the problem. The electronic circuitries realizing the mentioned computations are responsible for the high energy consumption. More specifically, their need for constantly being fed by electrons and more importantly the need for cooling. The latter contributes to the highest portion of the energy consumption. It is suggested that when photonic components are integrated into electronic circuitry, the computations are realized at a much lower energy cost. To simplify their claim, one can say that the advantage of the photonic components is thanks to the very nature of photons being different than electrons. Now let us listen to our guest Bogdan and what he has to say about this technology. Last but not the least we heard the gossips that that they are looking for a partner to launch a complete photonic miner demo!
Interview:
Can you tell our audience briefly what is a proof of work or proof of stake in the framework of crypto currency/block-chain transactions and crypto currency mining?
Proof of work is an algorithm that solves a particular problem: how to avoid double spending transactions in a decentralized ledger where no one trusts anyone. This algorithm is associated with a vast amount of computation performed during mining (writing transactions into blocks). The double-spending is prevented essentially because in a healthy network it is more profitable to behave as an honest actor. To better understand the jargon, please check the glossary. You may also want to check out this introductory post from the Mastering Bitcoin book.
In your paper you describe how high is the energy consumption of the current proof of work (done by electronic chips). How photonics can help in this regard?
From our perspective, the ideal proof of work algorithm would be an algorithm that spends no energy, but time (i.e. pure delay). This kind of proof of work is hardly possible in the real world. So instead, we would love to see some sort of a “computational delay” when only a small fraction of energy is spent on proof-of-work computation, while this computation itself introduces a time-delay. We believe that photonics can help us to achieve such a “computational delay”. The mentioned paper can be found here.
What are the physical principles behind the optical proof of work in your suggested devices. Are optical components able to perform the full computation of proof of work or do you use a hybrid system with electronics?
Our algorithm, Heavy Hash, consists of three stages: two SHA3 hashes and a matrix-vector multiplication in-between.
The photonic prototype we have is based on a mesh of Mach-Zehnder interferometers that optically performs matrix-vector multiplication. The other part of Heavy Hash, SHA3 blocks are computed digitally. Therefore, we have a hybrid system where most of the computational work has to be performed by the optical part (otherwise the energy reduction would be not significant).
How did you come up with this idea of optical proof of work, what are the previous works on the field that contributed to this? During your video lecture you mention neuromorphic computing and artificial intelligence studies helping you with your design, can you tell us about it?
Back in 2018 we have identified a huge need in lowering the energy use of Bitcoin. On the other hand there were multiple works on photonic computing such as by Paul Prucnal’s team, Marin Soljačić group and others. And me myself I did a PhD at FEMTO-ST in Laurent Larger’s group that worked on photonic reservoir computing. Later I met Mike Dubrovsky and we both thought that it was interesting to apply optical computers into alternative, non-AI use (i.e. for low-power proof-of-work mining).
We believe that proof-of-work is actually a simpler use case for optical accelerators (compared to AI) as you don’t need a huge capacity device to perform some useful computation. On the other hand, to be competitive in the field of machine learning, one needs to compute billions of multiply-accumulate operations per second. As an illustration, state of the art models in natural language processing tend to go up to hundred billions of parameters (e.g. GPT3). The same is true about computer vision and other machine learning fields.
Looking at the end result of your work, one can say that this combines multiple disciplines such as electronics engineering, computer science and physics. We, as Photonic Society Eindhoven, are mainly focused on electrical engineering and applied physics. Can you tell us about the computer science terms such as the hash rate, you refer to it very often, what does that simply mean?
The hash rate is simply the number of hashes that can be verified per second. This term is useful, for example, to compare two mining devices. The one that has a higher hash rate can faster find a new block. However, I would like to point out that it is even better to compare two devices using their hash rate per watt. As the higher hash rate per watt means a better profitability in general.
Eindhoven University of Technology is a strong player in integrated photonics with Institute of Photonic Integration and its cleanroom fabrication facilities. There is also an accumulation of know-how and expertise about design and fabrication of such devices. The material photonic components based on is InP. This semiconductor allows realization of on-chip lasers. We are wondering, what are your expectations from entities like us? On which parts of the project that you embarked on, you felt the need of assistance?
Indeed, even though the technology supporting Heavy Hash on the hardware level is already available, there are still many engineering challenges to be resolved (starting from how to minimize energy losses of individual components to the system as a whole, to improving optical proof of work in general). We are open for a collaboration with organizations like yours.
What would you consider as an improvement to your current design? For example, using multiple wavelengths at once can be attractive to you?
This is a very good guess: We are currently inspecting the use of multiple wavelengths.
You mention that high energy consumption of the current proof of work results in regional restrictions where these activities can be done. Such as regions with very cold environments or very cheap electricity. This is a negative aspect and causes instabilities. Do you think relying on photonic chips can make it inaccessible for certain regions?
We look for a scenario where a low-power mining chip is powered directly by a solar panel (or a similar renewable source). Having low-power mining chips can indeed open new opportunities.
Why create Heavy Hash, why not just compute SHA256 optically? This accomplishes the same goal without any changes to the Bitcoin codebase?
True, that would be nice to have SHA256 computed optically. However, the question is in efficiency. SHA256 is indeed computed very efficiently on digital computers. Therefore, there is no point implementing SHA256 optically if there is no guarantee that optical SHA will outperform digital computers in some foreseeable future. Maybe optical SHA will become viable in some ten years, however, we designed Heavy Hash to rely on technology already available today.
Because not all countries will have access to photonic fabrication facilities in case of a large demand. Do you think this can create new monopolies?
Our prototype is CMOS-compatible, therefore you can use older fabrication nodes. The ease of fabrication can potentially reduce the chance of new monopolies.
In terms of security of the proof of work system, can you foresee any malicious practices targeting optical proof of work easier than electronic version?
To reduce the change of malicious practices we already run a publicly accessible experimental oBTC (optical bitcoin) network (currently, a digital version only). We hope to get insights of the possibility of previously unseen attack vectors.
Contact: phe_hq@tue.nl, f.e.demirer@tue.nl.