Quantum Technology NEEDS Atomically Precise Manufacturing

Zyvex Labs has the goal of Atomically Precise Manufacturing (APM). They are researching and developing tools for creating quantum computers, analog quantum simulation devices and other transformational systems that require atomic precision. Developed as part of this effort, ZyVector™ turns the world-class ScientaOmicron VT-STM into an STM lithography tool, creating the only complete commercial solution for atomic precision lithography.

John N. Randall,

Today, we find ourselves in a similar situation, in that our manufacturing tools can make quantum computers but not (as of yet) good enough to outperform classical computers. However, the situation today is radically different than 20 years ago. Moore’s law is grinding to a halt, and we know that quantum computers will eventually outperform even the most powerful classical digital supercomputers in a number of very important applications that will have a huge impact on our national security. Also different is the international competition to advance computing power. The US is poised to achieve dominance in quantum technologies, but only if we invest in improving the technologies to manufacture them. The US Government should work with the private sector to achieve dominance in quantum technologies because it would be incredibly risky to our economic health and national security not to do so. One of the best ways to dominate the technology sector is to lead the development of the technologies needed to manufacture superior technologies.

In particular, in the area of quantum computing, none of the currently popular approaches to manufacture quantum computers will be the dominant quantum computing technology that will yield the dramatic capabilities that have been promised. I do not doubt the wisdom of well-funded attempts to make the best quantum computers that we can with the current manufacturing tools. This is imperative so that we learn what can work in the short term and build from there. The three most popular types of qubits are 1)Superconducting, 2)Ion Trap, and 3)Semiconductor Quantum Dot qubits. We must recognize that these approaches each have significantly limiting aspects that make them unlikely to be the scalable approach that will become the dominant quantum computing technology in the long run.

Let me explain why I believe this to be true.

1) Superconducting qubits’ microwave resonators are huge, on the order of a square mm with limited downscaling prospects. Superconducting qubits also have modest “coherence times” (which will limit their computing power), and while they do have a solid state component, a Josephson Junction device; it is simply a non-linear element in a microwave resonator.

2. Ion trap qubits are smaller than Superconducting qubits but still huge compared with the transistors that run our current digital, non-quantum computers. Both approaches may be compared to the vacuum tube technology that was the original digital computing technology.

Just as classical computers started out with non-solid state devices (vacuum tubes) but transitioned to integrated solid state devices, integrated solid state devices will also become the dominant technology for quantum computers. Superconducting and ion trap qubits can each be thought of as the vacuum tube technology that classical computers started out with.

!!!HEY WHAT ABOUT 3) Semiconductor Quantum Dot Qubits??!!!

The problem with Semiconductor quantum dot qubits is that it appears that they can be manufactured in today’s modern semiconductor fabrication facilities. One would think that current semiconductor manufacturing equipment which is currently producing solid state devices with ~10nm minimum features should be the obvious choice to make quantum computers. However, semiconductor fabrication tools have poor relative precision, on the order of ± 10%, which is acceptable in non-quantum digital computers, but insufficient for quantum devices. Extremely complex classical digital circuits can nevertheless be created with this technology, because classical bits only have to distinguish between 0 and 1 and just have to be on either side of a threshold. It is a testament to semiconductor engineers that they can make such complex systems with essentially really sloppy relative precision.

Let me put this in a context that most people can appreciate. Back when I was in school, during a housing boom in Houston Texas, I worked one summer building houses and apartments. Imagine I am up on a roof and call down to my buddy Zeke: “Zeke, cut me a 10 foot rafter!” and Zeke sends me a rafter +/- one foot! Imagine what that house would look like. Try to imagine making a car with that sort of relative manufacturing precision.

To understand the importance of manufacturing precision to quantum computing (and why they are so powerful), you only need to know that while they have digital inputs and outputs (1s and 0s) that internally they deal with a superposition of 1 and 0 states that allows them to represent a much larger range of possible solutions than either 1 or 0 while they are in the quantum state. Today’s digital computers have to double the number of transistors to double their compute power, while you only have to add one qubit to double the power of a quantum computer. How the quantum calculations go on is very complicated, but physicists have figured out how to map meaningful problems onto quantum processes and this realization has led to an international race with very high stakes.

In order to effectively harness the power of quantum computers, we can no longer live with the sloppy fabrication of today’s semiconductor factories. Maintaining the specific mixture of the superimposed quantum states is crucial to the successful completion of quantum computation which requires much more precision that is currently available in manufacturing tools. Also, keep in mind that quantum phenomenon is expressed most strongly at the atomic scale and even atomic scale variations in the fabricated physical dimensions of these devices will make computation more difficult. This situation results in much more stringent fabrication tolerances for solid-state quantum devices. I speak from experience having worked on solid-state quantum devices[1,2].

I reiterate my belief that it is entirely appropriate to make the best quantum computers possible with the manufacturing tools that we presently have. However, if we do not, as a nation, at the same time invest in developing manufacturing tools that have significantly better manufacturing precision, we may come out strong in the first quarter but lose the game. We may have short term success in building Noisy Intermediate Scale Quantum (NISQ) computers but fail to develop the competitive scalable universal quantum computers that we seek.
There are many paths to developing higher precision manufacturing tools. It would be prudent to fund a large portfolio of R&D efforts. What follows is not intended to be a pitch for our particular approach, as much as an indication that there are paths to follow to much higher manufacturing precision.

Quantum effects are typically exhibited at atomic and molecular scales and that therefore the most capable quantum devices will be manufactured with atomic-scale precision. There are a number of approaches to atomically precise manufacturing and several are applicable to quantum computers and other quantum devices. We are not alone in this belief. The DOE is funding a portfolio of atomically precise manufacturing programs. Work at NIST and Oak Ridge National Laboratories, are exploring atomically precise manufacturing for quantum and other technologies.

One method that my company has commercialized and is further developing is a technology, referred to as hydrogen depassivation lithography (HDL). It is a next-generation form of e-beam lithography that is carried out with a Scanning Tunneling Microscope (STM). The technical details are available in the scientific literature[3], but let me point out graphically just how much more precise at patterning HDL is than the best conventional e-beam lithography can do. The graph compares the normalized radial distribution of conventional e-beam lithography with that of HDL. The data for the conventional e-beam lithography is taken from an excellent paper by Karl Berggren of MIT[4]. I note that the conventional e-beam lithography distribution of dose must go out almost 4nm radially before the energy density drops to 10% of the maximum. With HDL the effective dose to expose drops 8 orders of magnitude at a radial distance of 0.5nm. HDL is a much sharper exposure tool. Sharp enough to do atomically precise patterning[5], and while it has other uses in solid-state quantum devices, it is being used to make single donor spin qubits[6]. It is also amenable to scaling up through massive parallelism that is simply not possible with conventional e-beam lithography[3]. Another approach being pursued at Oak Ridge National Laboratories is demonstrating atomic precision manipulation of matter with Scanning Transmission Electron Microscopes[7].

The US Government must invest in the future of our national security by funding research and development of an entirely new generation of manufacturing tools capable of atomic precision. I believe that this will be essential to achieve U.S. Dominance in quantum computing and many other valuable quantum technologies. I would be happy to provide many other details about the possibilities.

Reed, M. A., Randall, J. N., Aggarwal, R. J., Matyi, R. J., Moore, T. M., & Wetsel, A. E. (1988). Observation of discrete electronic states in a zero-dimensional semiconductor nanostructure. Physical Review Letters, 60(6), 535–537.

Broekaert, T. P. E., Randall, J. N., Beam III, E. A., Jovanovic, D., Seabaugh, A. C., & Smith, B. D. (1996). Functional InP/InGaAs lateral double barrier heterostructure resonant tunneling diodes by using etch and regrowth. Applied Physics Letters, 69(13), 1918–1920. https://doi.org/10.1063/1.117621

Manfrinato, V. R., Wen, J., Zhang, L., Yang, Y., Hobbs, R. G., Baker, B., … Berggren, K. K. (2014). Determining the resolution limits of electron-beam lithography: Direct measurement of the point-spread function. Nano Letters, 14(8), 4406–4412 https://doi.org/10.1021/nl5013773

Chen, S., Xu, H., Goh, K. E. J., Liu, L., & Randall, J. N. (2012). Patterning of sub-1 nm dangling-bond lines with atomic precision alignment on H:Si(100) surface at room temperature. Nanotechnology, 23(27), 275301 https://doi.org/10.1088/0957-4484/23/27/275301

Randall, J. N., Owen, J. H. G., Lake, J., Saini, R., Fuchs, E., Mahdavi, M., … Schaefer, B. C. (2018). Highly parallel scanning tunneling microscope based hydrogen depassivation lithography. JVSTB, 36, 6–10. https://doi.org/10.1116/1.5047939

Hill, C. D., Peretz, E., Hile, S. J., House, M. G., Fuechsle, M., Rogge, S., … Hollenberg, L. C. L. (2015). A surface code quantum computer in silicon. Science Advances, 1(9), e1500707–e1500707. https://doi.org/10.1126/sciadv.1500707

Kalinin, S. V., Borisevich, A. & Jesse, S. (2016). Fire up the atom forge. Nature, 539(7630), 485–487. https://doi.org/10.1038/539485a

20 thoughts on “Quantum Technology NEEDS Atomically Precise Manufacturing”

  1. Jan,

    You are correct that even in the massively parallel system described in article 5 that this would not be able to do classical digital consumer electronics. Classical digital electronics doesn’t need atomic precision. Solid state quantum computing does and does not need the throughput of today’s semiconductor fabrication facilities to make a major impact.

    In terms of tip longevity, we are making good progress. Tajaddodianfar, F., et.al. https://doi.org/10.1109/TCST.2018.2844781

  2. Jan,
    You pose excellent questions. This is a critical challenge that must be overcome if this is technology is to develop into a manufacturing technique. Indeed even without tip crashes the tip is subjected to very strong fields in the lithography mode. We know that tungsten atoms will exhibit surface mobility moving to lower surface energy configurations with no applied fields. This was first seen by Kellog via Field Ion Microscope imaging decades ago. We have also observed this motion via TEM imaging. It is a small miracle that tungsten STM tips are as stable as they are. With the improved control system though, we have seen remarkable tip reliability, more than adequate for research, but not yet enough for manufacturing.

    That is why we are exploring covalently bonded tips of very hard materials. We expect to achieve extremely stable tip structures.



  3. About the tip longevity. If you don’t plan to make STM-lithography over the whole wafer but only a small fraction of it, then you may not need millions of hours of MTBF (mean time between failure). But how long would a tip need to last in order to make a practical production system?

    I also wonder if the tip is worn even if it not crashed into the surface? Even though the tip bias is low when making the fine lithography (~2 V?), this should still produce some electrostatic force on the single apex atom. It this atom is moved or ejected, the tip would have to be recalibrated or exchanged. By recalibrated, I mean that the apex would have to be located relative to the amplification settings by scanning some known structure and add offsets to the scanning movement.

    So, is the scanning tip “worn” by the STM-scanning even in the absence of tip crashes, at the level of thousands of hours of MBTF?

  4. How many qubits do you need to make in order to make a useful chip? The proposed system has a fine scanning area of 100nm x 100 nm and it takes about 6 minutes to expose this area. Presumably, you would have to make not only the qubit with the STM lithography, but also some part of the leads (“fine leads”).

    A conventional lithography step would then make the (coarse) leads as to connect to the qubits, but since this step has a limited resolution (and line width variation), the “fine leads” need to extend some distance away from the actual qubits. I.e. you will have to “waste” some of the atomically resolved lithography on making leads.

    Could you make anything useful in an area that is a small multiple of 100 nm x 100 nm?

  5. Interesting. Two questions. Is there a publication with the method that you describe? It would be interesting to see just how close to atomically perfect production you can reach with the method.

    Second question. Have anyone fabricated a device with the method? Even though single electron transistors is not a viable computing element for a classical computer, if you could make ten out of ten room temperature single electron transistors, you would have demonstrated the ability to make sub 10-nm structures repeatably. Likewise, a high yield production of sub 10 nm (classical) transistors would be an indication of the methods usefulness.

  6. Jan,

    In terms of science, there has been a wealth of science done in pursuit of quantum computing. Most of this has been published by the Michelle Simmons group at the University of New South Wales in Australia.

    We are working with NIST and others to make analog quantum simulation devices with this technology that are designed 2D arrays of dopant atoms that will complement the array of cold atoms that are being used to better understand condensed matter quantum physics.

    Thank you for your comments.



  7. Jan,

    Forgive me for not being more clear. The STM lithography that we describe is able to place dopant atoms in a single buried (100) plane of silicon. The STM lithography removes H atoms from a Si surface (not metal) and a gas (PH3) is introduced into the chamber. The PH3 sticks where the H isn’t in a self limiting fashion. All of this (up until this point) is done at room temperature.

    A brief anneal has the P atoms exchange with surface Si atoms. Careful low temperature (250C) epitaxy grows a few tens of nanometers of Si with little if any diffusion of Si atoms.

    Qubits for quantum computing are not yet the exciting (practical) technology that they will be, but there is a very well funded global race that is betting that quantum computing will be game changer.



  8. “In terms of the technology being at a state where it is only good enough for making some relatively small research devices, you are indeed correct. However, that has been the path for every useful lithography technology. avs.scitation.org”

    Let’s hope you can make some good science with these methods..

  9. I forgot to than you for your answer! It is very nice when an author takes the time to reply to reader questions; you deserve a thanks for this!

    About you points..

    “In terms of what you can do with a pattern of H atoms on Si, article [6] points to the currently most useful pattern transfer for quantum devices, putting dopant atoms in a single buried 100 plane of a Si crystal.”

    The cited article is a theoretical article about how to make devices with dopant atoms. But your method is about removing hydrogen atoms from a metal surface. So, in order to use this pattern, it would have to be transferred to a Si wafer (by imprint methods?). This transfer is unlikely to be perfect (we can compare with the difficulties that the imprint lithography community has had) but even if it were, the standard way to implant dopants is by applying heat. And diffusion implantation of the transferred pattern will be completely “washed out”, since the thermal motion is isotropic and random…

    “Many devices such as single electron transistors and donor spin qubits have already been made with this patterning technology. ”

    The cubits may be valuable, I don’t know. Single electron transistors have been proven to be not useful for computers. This is true even if an ideal single electron transistor is considered….

  10. “STM’s are notoriously slow. However, article [5] describes a path to massive parallelism with current technology. ”

    Article [5] is not that bad. However, the authors estimate the maximum areal throughput of their “dream” system to less than 10^6 um^2 per hour. Scanning a complete 300 mm wafer would take 8 years of around the clock exposure… Note, to reach this throughput they would have to solve the hereto unsolved problem of tip longevity. It would have to be increased from of-the-order-of-hours to millions of hours…

  11. thanks for the reply JNR,

    Eric Drexler once said the first thing he’d make when the assembler is built is a computer. Looks like this has turned prophetic. Seems the first thing one can make is a quantum computer!

  12. Dear Flash,

    We have many problems that we are working hard to resolve. One of them is money. We do have 4 Government research contracts that are helping with funding, but we are also actively seeking investors to help us meet the opportunity of selling into the rapidly growing number of quantum research programs largely driven by the signing of the National Quantum Initiative that has put 1.275B$ into quantum research over the next 5 years.

    Thanks for asking.


  13. Dear Jan Jansson,

    Your objections are understandable.

    STM’s are notoriously slow. However, article [5] describes a path to massive parallelism with current technology.

    In terms of what you can do with a pattern of H atoms on Si, article [6] points to the currently most useful pattern transfer for quantum devices, putting dopant atoms in a single buried 100 plane of a Si crystal. Many devices such as single electron transistors and donor spin qubits have already been made with this patterning technology. There is also published work on selective Atomic layer deposition of metal oxides which have been used as a reactive ion etch mask to transfer the pattern to Si https://doi.org/10.1116/1.4890484

    In terms of the technology being at a state where it is only good enough for making some relatively small research devices, you are indeed correct. However, that has been the path for every useful lithography technology.

  14. Now I have perused the number [4] article.. and I am not that impressed. Sure, they draw squiggly lines with sub 1-nm resolution, but its only removing hydrogen atoms from a Si surface. There aint nothing you can do with that. No mask, no functionality, no devices..

  15. OK, even if the STM-based lithography would be much better in terms of resolution – doesn’t surprise me – what about throughput? STM’s are *notoriously* slow. You would be lucky to hit 1000 pixels per second.

    Here is an article that describes a “fast STM” [1]. They have achieved 5 mm per second scanning and 200 kHz feed-back.. So this STM lithography seems to be a tool for making single devices for science experiments. Wake me up when you have something reasonable fast…


  16. Brian, there seems to be a type in you text:

    “!!!HEY WHAT ABOUT 3) Semiconductor Quantum Dot Qubits??!!!”

Comments are closed.