August 23, 2017 09:00 PM GMT Global Technology Quantum computing weird science or the next computing revolution? With more companies moving quantum computers from the lab to commercial activities, we believe widespread quantum computing is about to become a reality and holds the key to double the high-end computing market from $5bn to $10bn. Morgan Stanley does and seeks to do business with companies covered in Morgan Stanley Research. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of Morgan Stanley Research. Investors should consider Morgan Stanley Research as only a single factor in making their investment decision. For analyst certification and other important disclosures, refer to the Disclosure Section, located at the end of this report. + = Analysts employed by non-u.s. affiliates are not registered with FINRA, may not be associated persons of the member and may not be subject to NASD/NYSE restrictions on communications with a subject company, public appearances and trading securities held by a research analyst account.
Global Technology Quantum computing weird science or the next computing revolution? Quantum computing is at an inflection point - moving from funda- credible internal quantum computing roadmaps are IBM, Google, mental theoretical research to an engineering development phase, Microsoft and Nokia Bell Labs. We also believe that within the dis- including commercial experiments. That said, in the medium term, we ruption path there is room for new companies to emerge as impor- see a transition period during which classical computers will simulate tant players, such as D-Wave and Rigetti. Outside of tech, companies quantum algorithms, while genuine quantum computers are custom- having led the charge for several years include Airbus, Lockheed ised to fit those algorithms. Martin, and Raytheon in the aerospace and defence sector, with While the classical computer is very good at calculus, the recent interest by Amgen and Biogen (for molecule simulation) and quantum computer is even better at sorting, finding prime num- Volkswagen (traffic optimisation). bers, simulating molecules, and optimisation, and thus could Life beyond transistors? Because quantum computing is not open the door to a new computing era. Moore's Law was the main suited to all compute tasks, smartphones, PCs, and web servers driver of the digital revolution; we believe quantum computing could storing data will continue to run on current technology. We think trigger the beginning of a fourth industrial revolution, with far- the high-end compute platforms could see a transition post 2025, reaching consequences for many sectors where computing power is similar to how steam engines coexisted with combustion engines and becoming a limitation for R&D, such as Financials, Pharma (drug dis- electric motors for decades before being decommissioned. In the covery), Oil & Gas (well data analysis), Utilities (nuclear fusion), medium term, we see incremental demand for FPGAs and GPUs (pos- Chemicals (polymer design), Aerospace & Defense (plane design), sibly benefiting Xilinx, nvidia, and maybe Intel) as more supercom- Capital Goods (digital manufacturing and predictive maintenance), puters from Atos and Fujitsu are developed to simulate the Artificial Intelligence, and Big Data search in general. behaviour of quantum computers. If quantum computers eventually It is hard to discern among currently evolving hardware platforms, do become ubiquitous, then the growth of high-end computing sys- which will be resilient enough to beat classical supercomputers in the tems that emulate them could be affected, hence limiting the valua- next few years, but in our view the listed companies with the most tions of those stocks, but this is more a post 2020 event, in our view.
Contributors Morgan Stanley & Co. International plc Morgan Stanley & Co. LLC Morgan Stanley & Co. LLC Francois A Meunier Katy L. Huberty, CFA Keith Weiss, CFA Equity Analyst Equity Analyst Equity Analyst +4420 7425-6603 +1212 761-6249 +1212 761-4149 Francois.Meunier@morganstanley.com Kathryn.Huberty@morganstanley.com Keith.Weiss@morganstanley.com Morgan Stanley & Co. LLC Morgan Stanley & Co. LLC Morgan Stanley & Co. LLC Joseph Moore Brian Nowak, CFA James E Faucette Equity Analyst Equity Analyst Equity Analyst +1212 761-7516 +1212 761-3365 +1212 296-5771 Joseph.Moore@morganstanley.com Brian.Nowak@morganstanley.com James.Faucette@morganstanley.com Morgan Stanley & Co. International plc Morgan Stanley & Co. LLC Morgan Stanley & Co. LLC Adam Wood Vinayak Rao, CFA Elizabeth Elliott, CFA Equity Analyst Research Associate Research Associate +4420 7425-4450 +1212 761-4669 +1212 761-3632 Adam.Wood@morganstanley.com Vinayak.Rao@morganstanley.com Elizabeth.Elliott@morganstanley.com Morgan Stanley & Co. LLC Erik W Woodring Research Associate +1212 296-8083 Erik.Woodring@morganstanley.com
Contents 5 Executive summary 11 The world needs more (affordable) computing power disruption needed 14 Price reduction in classic computing has potentially run its course the quantum computer is not a "nice to have" but a "must have" for global productivity 15 Potential $10bn addressable market for quantum computing 16 Quantum computing moving to the engineering development phase 25 What is the potential impact of quantum computing to the stock market and the economy? 27 Glossary of Terms 4
Executive summary Quantum Computing at a Glance What is quantum computing? While classical computers use the laws of mathematics, quantum computers use the laws of physics. We can describe the difference between classical computing and quantum computing with the image of a coin. In classical computing, information is stored in bits with two states, 0 or 1 or heads or tails. In quantum computing, information is stored in quantum bits ("qubits") that can be any state between 0 and 1 similar to a spinning coin that can be both heads and tails at the same time. The information contained in qubit is much richer than the information contained in a classical bit. Every time you look at a spinning coin it takes on a different state, i.e. 75% heads and 25% tails, depending on where in the rotation you look at it. This is similar to a qubit that has a very fragile state, which can change each time you look at it. To search potential solutions to a problem, a classical computer individually tests different combinations of 0s and 1s. Meanwhile, quantum computing can test all combinations at once because a qubit represents all combinations of states between 0 and 1 at the same time. In a quantum computer, qubits are interconnected by logic gates, like in a classical computer, but the available operands are more diverse and more complicated. What is the holy grail of quantum computing? Exponential acceleration. In other words, a quantum computer would be able to compute at a much faster speed (exponentially faster) than a classical computer. This implies that classical algorithms, which would take years to solve on a current supercomputer, could take just hours or minutes on a quantum computer. What are the hurdles to achieve quantum computing? Qubits are fragile and the state of a qubit can always change, therefore a significant challenge in quantum computing is maintaining a stable state of qubits in order to read the information. Quantum computers must operate at extremely low temperatures to slow the movement of qubits and are much larger in size than classical computers. Maintaining a stable state of qubits and correcting for the "noise" of high error rates as the number of qubits scale are some of the largest hurdles in quantum computing, which companies are solving in different ways. It is also difficult to read the result of an experiment without damaging the quantum state of a qubit. As a result, the read-out of a quantum experiment is not always the same - it gives a "probabilistic" answer, while a classical computer gives a deterministic' "answer (always the same and predictable). What are the use cases? In order to be practical, quantum computers should have at least >50 qubits, relative to 16 [units?] today, and with a lower level of noise. Early applications are expected to be used in the chemistry and pharmaceutical industries, where scientists can decrease the time to discover new materials, for example new catalysts to produce fertilizer with much less energy than today. With the addition of more qubits, applications are expected to grow into the finance (i.e. portfolio optimization) and machine learning spaces. What is the time line? Google, IBM, and others have achieved 16+ qubit systems, while others have simulated quantum computing with more than 40 qubits. That said, quantum computers are measured by not only the number of qubits but also by the amount of interconnection (the more interconnection, the more flexible) and the noise level (the lower the better). Google is tight-lipped about its efforts, but IBM has talked about having a 50-qubit system within the next few years. We believe that there will be a two- to three-year period during which simulations and real quantum hardware systems will coexist before quantum computers reach more than 50 qubits and simulators become less relevant. Beyond that, it is all about "scaling" the number of qubits at an acceptable level of noise, similar to what the computer/semiconductor industry experienced by doubling the number of transistor every 18 to 24 months for the same production cost (which is known as Moore's Law). Morgan Stanley Research 5
Quantum computing is at an inflection point Moving from fun- Milestones investors should watch for to gain interest in the damental theoretical research to an engineering development phase quantum computing investment theme: to commercial experiments. No quantum computer today can beat a super computer and there is no certainty around who has the best 1. One or more companies builds a quantum computer capable hardware roadmap into the next 10 years, but the recent progress of beating any classical computer, but without a specific com- shows a significant improvement in terms of the algorithms that can mercial purpose. be run more economically on quantum computers than on traditional computers. 2. One or more companies builds a quantum computer capable of beating any classical supercomputer, with a specific commercial purpose We believe that we in are a transition period We believe that the 3. One or more companies launches a new product or service core of the quantum computer, the qubit, is ready, be it in supracon- that could only have been designed on a quantum computer, ducting, trapped ion or even compound semiconductor form. The and as a result sales and revenues from quantum computer current engineering phase is all about the scaling of those qubits and companies (public or private) reach a significant level and their interconnection. While this engineering phase is underway, we make the incremental market size of >$5bn credible. believe we are in a transition period until physical quantum computers reach 50 qubits (from less than 20 today) with classical computers simulating the behaviour of quantum computers in order to expand the number of quantum-specific algorithms until they are capable of beating classical computers (which is known as "exponential speedup"). Exhibit 1: Quantum Computing potential timeline Source: Morgan Stanley Research 6
The world economy needs more computer power, but poten- Quantum computing does not make the classic computer irrelevant tially of a different kind. We see an increasing need for supercom- smartphones and laptops will still use transistors for the foresee- puting power amongst many verticals (financials, chemicals, oil & able future and the transition might take several years. As explained gas, pharma, nuclear fusion) and AI/Big Data search in general. As the in our digital revolution blue paper (Technology: Disruptions and pro- digitalization of the economy continues, the need for a different form ductivity growth in the next decade of the digital revolution), of computing is accelerating. More and more devices and systems are quantum computing could be a key driving element of a fourth indus- being connected to collect data (the Internet of Things),. Artificial trial revolution as it could, for example, help invent new molecules intelligence chips are finding correlation levels or making inferences for drugs and materials that could not be invented before with tradi- in large quantities of data but there is also enough evidence to sug- tional computers. gest that large companies are looking for larger levels of computing power, or computing of a different kind. The classical computer is a large calculator that is very good at calculus and step-by-step analytics, whereas the quantum computer looks at solving problems from a higher point of view. Exhibit 2: Comparing classical computers with quantum computers at a high level Source: Morgan Stanley Research Morgan Stanley Research 7
Exhibit 3: Quantum Computing in the Industrial Revolution time scale Source: Morgan Stanley Research Quantum computing opens up new markets for computing. It may Exhibit 4: not be able to solve everything but it could solve issues that conven- Classical computers are great at calculus, quantum com- tional computers either cannot, or struggle to solve. puters could open new horizons Source: Morgan Stanley Research 8
How big could Quantum Computing be? In terms of landscape, in the exhibit below, we divide the technology road maps into three categories: The market for high end computing is $5-6bn a year, according to IBM, and according to our discussions could increase to $10bn with poten- 1) Companies looking to build universal quantum computers. There tial upside should quantum computing be used for many algorithms are three main subcategories, based on the material/technique used in many industries. The size of the market will also depend on the to create the qubits, i.e. supraconducting, ions trapped with lasers, business model used (one-off hardware sales vs. cloud-based, the and compound semiconductors. The three road maps are explained latter being the most likely, in our view, as the hardware needs to run in more detail later in this report. at a very low temperature (below 1 degree Kelvin) in a very stable radio frequency environment, and as such is more likely used as a 2) Companies which have built a task-specific quantum computer - shared asset). D-Wave and its partners in particular. While the level of noise is higher than average and the level of connectivity is lower than for Exhibit 5: universal quantum computers, D-Wave has generated revenues for How big could quantum computing be? several years with task-specific computers containing up to 2,048 qubits. 3) Companies offering simulating platforms for quantum computers, i.e. classical computers able to simulate quantum bits and quantum gates, with varying degrees of accuracy. The largest simulators could simulate quantum computers with a number of quantum bits in the high 40s. The best simulators can simulate accurate levels of noise. There are also software companies working on quantum algorithms that could be used on any platform and are currently designed using simulators. We have also identified a few suppliers of key hardware components to the nascent quantum computing industry such as cryogenics equipment (as most hardware road maps need to "freeze" the qubits at close to absolute zero) and radiofrequency equipment (a signal at several gigahertz is used to maintain the qubit in a quantum state). Source: Morgan Stanley Research estimates Our discussions with industry specialists at or involved with IBM, D-Wave, Lockheed Martin, Microsoft, Rigetti, BluFors, Google, Intel, Atos, 1Qbit, Nokia Bell Labs and Morgan Stanley IT, show that quantum computing is moving from white papers and fundamental research to engineering developments, including roadmaps to full functioning quantum computers with hundreds or thousands of qubits. The range of effort varies by company - from several key scientists at the Bell Labs to hundreds of engineers at IBM. According to The Economist, there is $1,500bn of public funding dedicated to quantum computing in the next several years. Morgan Stanley Research 9
Exhibit 6: Quantum computing technology and company landscape Source: Morgan Stanley Research 10
The world needs more (affordable) computing power disruption needed With an abundance of computing power in our pockets, it does not Exhibit 7: feel like more computing power is needed. That said, this abundance Quantum computing has a much larger reach than a classic is only related to personal productivity and we are gathering enough computer and thus a much larger potential addressable evidence to convince us that several large companies and sectors are market, in our view looking for more computing power at an affordable price. Moore's Law is becoming more difficult to achieve, i.e. the cost per transistor does not halve every 18 months as it used to. ASML's EUV is expected to bring some relief for the next 5 years but beyond that the size of the transistor is getting so small that Moore's Law is likely to reach a limit. The classic computer architecture cannot solve all problems. The classic computer architecture should be viewed as a large calculator, e.g. very good at calculus ; however there are problems that are not naturally suited for a classic computer's architecture. Hence, we have seen the emergence of more specialized chips for graphics (GPUs), and specific calculations (DSPs, TCUs, FPGAs). But there are numerous tasks where quantum computers can thrive naturally, such as the factoring of large numbers, sorting numbers, artificial intelligence, solving optimization problems such as supply chain Source: Morgan Stanley Research optimization, and vehicle routing problems as well as the simulation of quantum mechanics. Exhibit 8: While quantum computers are still mostly in the lab and slowly A quantum computer is much more efficient at sorting numbers than moving to the cloud, companies and scientists are looking for a classic computer for a range of applications ("exponential speedup") evidence that quantum computers can beat classical computers: The number of computing steps required to sort a large list of numbers is exponentially higher for a classic computer than for a quantum com- Quantum computing could prove more efficient than today's com- puter puters for sorting and searching. Today's best algorithms for sorting 1.E+20 n numbers can achieve a full result in n.log(n) steps while a quantum 1.E+18 computing cans sort in a square root of n, i.e. it takes many more 1.E+16 1.E+14 computing steps for the classic computer to solve the same problem. 1.E+12 Moore's Law has allowed those computing steps to become signifi- 1.E+10 1.E+08 cantly quicker and cheaper but with Moore's Law slowing, the bene- 1.E+06 fits of quantum computing become more apparent. To sort a billion 1.E+04 numbers, a quantum computer would require 3.5 million fewer com- Classic Computer Quantum Computer 1.E+02 1.E+00 puting steps than a traditional computer and would find the solution in only 31,623 steps. This gain would be particularly interesting for companies involved in databases, big data analytics and Source: Morgan Stanley Research estimates search engines. Morgan Stanley Research 11
In Financial Services, discussions with quantum computing soft- with IBM show simulation of molecules with 5 atoms already, but this ware company 1Qbit show that the most promising applications is something which can be done with a laptop computer as well. The could be in: 1) collateral optimization (managing stock of collateral to promise of quantum computers is the simulation of much larger mol- optimize profits); and 2) portfolio construction optimization con- ecules which cannot be simulated by a supercomputer (above 50 strained by transaction cost. For the moment, 1Qbit suggests the atoms), represented by 5 qubits. best available hardware platform is actually based on traditional transistors (a cluster of FPGA chips simulating a quantum computer) In Pharma, Biogen announced in June 2017 a collaboration with but when quantum computing power is high enough, the software quantum computing software company 1QBit and Accenture to would switch to the new generation hardware, potentially custom- design new molecules. Morgan Stanley analyst Matthew Harrison ised to those specific algorithms for financial services. Goldman believes Amgen is also working on quantum computing experiments. Sachs is an early investor in D-Wave. As with chemicals, the quantum computers appears very well suited to simulate a whole molecule. In Chemicals, Paul Walsh and team believes this is the first time that a large company like BASF has announced a plan to set up In Oil & Gas, Martijn Rats and team have identified BP as a strong a supercomputer in order to design polymers with pre-defined user of computing power to optimize oil extraction. Today, it holds properties. The goal for BASF is to simulate the quantum mechanics more than 1 petabyte - a billion million bytes - of data. ARGUS, the and therefore the properties of a molecule - while today the simula- wells data platform of BP, contains 2,458 wells or 99.5% of all BP s tions are made with a traditional cluster of computers, we believe well stock. that quantum computers are more naturally able to solve such problems. BASF and HewlettPackard announced that the companies will APEX is the system optimizer. It enables BP to optimise production collaborate to develop one of the world s largest supercomputers in real time by monitoring and modelling physical constraints across for industrial chemical research at BASF's Ludwigshafen headquar- the production system live. It will be installed in all operated assets ters this year. Based on the latest generation of HPE Apollo 6000 by 2018. systems, themselves based on Intel Xeon chips, the new supercomputer will drive the digitalization of BASF's worldwide research. The SIRAAJ is BP s field development platform it was pioneered in new supercomputer will promote the application and development of Oman. It enables live updating of the development plan as results complex modeling and simulation approaches, opening up completely stream in from new drilling data. BP can now monitor and analyse new avenues for our research at BASF, according to Dr. Martin Brud- drilling operations and wells, from remote centres around the world. ermueller, Vice Chairman of the Board of Executive Directors and BP highlights that this will improve its decision making and make the Chief Technology Officer at BASF (press release by BASF published organisation more efficient. on the 17th March 2017). The supercomputer was designed and developed jointly by experts from HPE and BASF to precisely meet our Another system, The Plant Operations Advisor, is in pilot testing on needs. The new system will make it possible to answer complex a facility in the Gulf of Mexico. At the moment it is running over 20 questions and reduce the time required to obtain results from sev- million calculations per day and assesses the operational state of 150 eral months to days across all research areas. As part of BASF s digi- pieces of equipment, every 2 seconds. By end 2018, BP expects to talization strategy, the company plans to significantly expand its have it fully deployed. capabilities to run virtual experiments with the supercomputer. It will help BASF reduce time to market and costs by, for example, simu- Eni fired up its new HPC3 (High Performance Computing) in the lating processes on catalyst surfaces more precisely or accelerating Green Data Center in Ferrera Erbognone (PV). HPC3 allows Eni to the design of new polymers with pre-defined properties. fully support all the activities in the Exploration and Production sector. The High Performance Computing HPC3, together with the We believe that quantum computing could help in the Chemicals coexisting HPC2 system, will provide Eni with a sustained 5.8 Peta- industry with the design of new chemicals and materials, and Dow FLOPS, and 8.4 PetaFLOPS of peak computing capacity. Chemicals announced a collaboration with quantum computing software company 1Qbit on June 21, 2017, in order to create new The new cluster continues along Eni s HPC philosophy based on molecules. A quantum computer is naturally suited to simulate the hybrid architectures, by using top end GP-GPUs as computational quantum physics linking atoms in a molecule (i.e. Schroedinger's accelerators. Driven by Eni s internal research activity, the cluster equation) especially as the number of atoms grows. Our discussion design targets both the most efficient energy solution and the 12
delivery of the maximum computational power required by the most taining an unknown string of bits, they showed that just a few super- advanced proprietary algorithms. HPC3 records a remarkable energy conducting qubits can discover the hidden string faster and more efficiency consumption of 3.66 gigaflops/watt; moreover, overall efficiently than today s computers. Their research was published in efficiency is also maximized by the direct free-cooling solution pro- the paper, Demonstration of quantum advantage in machine vided by the hosting Eni Green Data Center. learning (Nature article published on the 13th April 2017) HPC3 is an intermediate step towards the next evolution, the HPC4, According to a report by CBInsight, Airbus Group established a team expected at the beginning of 2018. With HPC4, Eni s target is to over- to tackle quantum computing at the end of 2015 at its site in New- come the barrier of 10 PetaFLOPS. port, Wales. Airbus wants to adapt existing quantum machines to specific problems within the aerospace industry, namely those In aerospace and defence, companies have been early to see the requiring the handling and storage of large amounts of data, advantages of quantum computing with Lockheed Martin and NASA including sorting and analyzing images streamed by satellites, or cre- being early partners of D-Wave (since 2010 for Lockheed) and Ray- ating ultra-durable materials for aircrafts. theon with IBM. In transportation, Volkswagen announced a partnership in March In May 2017, Scientists from IBM Research and Raytheon BBN demon- 2017 with D-Wave to optimize traffic flow. The constant quest for strated one of the first proven examples of a quantum computer s more computing at an affordable cost sets the scene for a potential advantage over a classical computer. By probing a black box con- technology disruption. Morgan Stanley Research 13
Price reduction in classic computing has potentially run its course - the quantum computer is not a "nice to have" but a "must have" for global productivity We believe one of the main drivers of each of the three industrial With only incremental cost reduction available from classic tran- revolutions is a cost curve. The first industrial revolution was driven sistor based computers, the tech industry has found a way to put by an increase in efficiency for the steam engine, hence the cost of resources in common in the form of public cloud and thus lower com- producing mechanical force was reduced by a factor of 5 over time. puting cost at a higher rate than Moore's Law would allow in the past The same occurred during the second industrial revolution when the decade. (Global Technology: Public Cloud, What s It Worth?, Semi- cost of producing mechanical force was reduced by a factor of 14 over conductors: Making the case for Intel to lower its data center targets time. Moore's Law drove the cost of computing power down by a ) factor of 100 million during the digital revolution. That said, the difficulty in driving Moore's Law further in terms of cost reduction is well Once a large part of the globally available computing power is in documented (Blue Paper: Global Semiconductors: Chipping Away at public clouds, it is difficult to lower computing power significantly Returns,Technology - Semiconductors: Global Insight: Moore's law further, hence the need to work harder on quantum computing tech- slowdown temporary or terminal? and could be summarized by the nologies, which have actually been well documented for decades but following exhibit. have not moved to the engineering phase yet. Exhibit 9: Classical computers do not scale as fast as they have in the past Source: Morgan Stanley Research 14
Potential $10bn addressable market for quantum computing The high-end computing market is worth around $5bn today. It is We expect the next 10 years to see a rebalancing favoring commercial dwarfed by the overall commercial and consumer computing market. vs consumer computing as quantum computers are able to solve The consumer computing market is 3x as large as the commercial many problems that could not be solved up until now. Assuming the computing market as the past 10 years have seen smartphones ratio of consumer vs commercial spending remains the same, then emerging as the leading computing platform (the iphone was the high end computing market, including quantum computing, launched in 2007). IBM thinks quantum computing can target the could comfortably reach $10bn in the next ten years, in our view. Any high end computing market and potentially double it as commercial rebalancing of spending, between consumer and commercial com- applications emerge. We believe that the potential could be larger puting could lead to further upside to that $10bn revenue forecast. when quantum computing starts scaling like classical computers scaled from 1980 onwards. Exhibit 10: Exhibit 11: How big could quantum computing be? The market for high end computing including Quantum Computing could double in the next ten years $bn smartphones tablets PCs consumer total consumer computing PCs commercial Servers High End and Quantum Computing total commercial computing consumer vs commercial computing 2017 415 50 125 590 125 60 5 190 3.1 2027 872 185 89 10 283 3.1 Source: Morgan Stanley Research estimates Source: Morgan Stanley Research estimates Morgan Stanley Research 15
Quantum computing moving to the engineering development phase Quantum computers have been in the lab for several decades. punch cards to touch screens, and this is one of the challenges today British physicist Brian David Josephson predicted in 1962 that a cur- for quantum computers. The first idea of a quantum computer rent could flow without any voltage applied to a superconductor-in- emerged in 1981 at the Physics of Computation Conference at the sulator-superconductor junction, while a classic semiconductor MIT, when Nobel Prize winner Dr. Richard Feynman showed that a transistor needs voltage for current to flow through it. Josephson classical computer (a Turing machine) would experience an exponen- won the Nobel Prize in Physics for this discovery in 1973 and the junc- tial slowdown when simulating quantum phenomena, while his tion, which bears his name, is one of the two methods used today to hypothetical universal quantum simulator would not. Even the create the basic unit in a quantum computer. In a classic computer, largest, most powerful computers available today would never be the basic unit is a bit, made of a on or off transistor. In a quantum able to simulate a quantum system with 50 qubit in a reasonable computer, the basis unit is quantum bit or qubit. That bit has many amount of time. No quantum computer today can achieve this either, shades of grey between 0 and 1 and can even be 1 and 0 at the same but the progress made in the past 12 months make us believe that this time. In other words, instead of being a clear state, true or false, a barrier is reachable for a few companies. qubit is a statistical representation of a constraint or state. Hence the complication to enter and output data from the computer. It took We summarize below the main differences between today's com- decades to simplify the input and output of classic computers, from puters and a quantum computer. Exhibit 12: Comparing the performance of classical computers and quantum computers at a high level Source: Morgan Stanley Research 16
Meanwhile, IBM has announced significant improvements, with 16-17 2) Scalability of the system as current qubits are fairly large, in the qubits now available on its cloud platform, ahead of schedule and range of 1um, which means that a chip with 1 million qubits with the with a roadmap to 50 qubits in the next few years and up to 1,000 current dimensions would require a million RF generators and their qubits longer-term. Google keeps its progress internal but a report associated multiplexers. That would fill an area as large as a football by futurism.com shows that Google is also working on a similar archi- pitch. tecture, but with a particular focus on reducing the number of "errors" made by the quantum computer up to 9 qubits (according to 3) The material used for the Josephson junction, which could be made the last Google quantum publication) and has been reported by of either the very exotic material Niobium or the very common mate- futurism.com to test a 20 qubit chip this year with a goal to achieve rial aluminum. 50 qubit by the end of 2017. Trapped ion roadmap, promising but still in Universities - Recent There are currently three different road maps to create a universal improvements have been made around a different way to make a computer with no established winners, but the progress made in the qubit-based computer, not with supraconductors, but with lasers past 12 months, at IBM for instance, is notable and makes us believe and electromagnetic fields, as described with the Paul ion trap. The that quantum computers could be at an inflection point. ion trap was conceived of in the 1950s by Wolfgang Paul, who was awarded the Nobel Prize in Physics in 1989, and it was first imple- Supraconductor roadmap, the most visible so far - The supracon- mented in 1995 by J. Cirac and P. Zoller. While this road map is inter- ducting route chosen by IBM, Google, D-Wave and Rigetti seems to esting in terms of its lower error rate, the speed of the computer is be making more progress at the moment, even if it is limited by the lower than with supraconductors and the lack of backing by large number of radio frequency drivers it can handle, and taking the corporates could be a headwind. The efforts for this road map are led system to a temperature below that of outer space is not an easy more at a University level (such as the University of West Sussex, task. That said, discussion with the main supplier of cryogenic England, and the University of Maryland, with a start-up spin-off devices, BluFors Cryogenics, show that a triple stage cooling equip- called IonQ). ment works well with enough radio frequency coaxial cables feeding to the core of the supraconducting chip. Also, the supraconducting A team of scientists, including Austin Fowler from Google Research, quantum computer can run 10,000x faster than the trapped ion- have published a blueprint for a quantum computer, with a modular based quantum computer, i.e. up to 100MHz, but it makes more mis- approach based on ion trapped qubits and solving many engineering takes. That is why the leaders in this road map are working to improve issues to build it and interconnect the different modules. We have not only the number qubits but also the error rate. seen the core of the quantum computer, with lasers and magnetic fields, at the University of West Sussex (Brighton, England) headed D-Wave has sold quantum computers since 2010 and IBM delivers its by Winfried Hensinger. Hensinger has enough funding to actually 5 and 16/17 qubit quantum computer as a free cloud service (in make a first >10 qubit prototype and is currently looking for $100m exchange for access to the code). Google, in our view, will keep its funding to build a larger scale prototype with over 1,000 qubits. research internal and it would make sense to stay that way if Google finds a way to lower the cost of sorting, searching, and machine While the core of the quantum computer seems to work correctly, learning with a quantum computer. the team of scientists have also solved several practical issues such as interconnecting the trapped ions qubits between each other. Inter- There are some clear positives about the supraconducting roadmap connection with photons used to be the limiting factor, with a max- - the qubits are well identified, with a good level of coherence and imum frequency of 7 Hertz (7 times per second) due to the interfacing fidelity with a road map to improve the error rate. between the trapped ions and the photons emitter/detector. The paper suggests to use ions, the same ones that are used in the ion There are, however, some outstanding issues: trap, as interconnection between the different modules, increasing the speed by a factor of 100,000 and thus reaching 700kHz. This 1) The need for very low temperature, below 1 Kelvin. But the sounds very low compared to a clock speed in most personal com- progress made by cryogenics companies like Blufors is impressive. puters around 3GHz, but each qubit can do so much more than a classic bit that the clock speed does not matter so much, according to Hensinger. Morgan Stanley Research 17
Traditional industrial chips will be used to control the magnetic fields temperature (i.e., colder than deep space) and could scale up with no that shift ions in and out of the traps - digital to analog converters need for hundreds of radio frequency generators to maintain the ion with 16 bit precision and 1MHz refresh rate, which could typically be in entanglement - but the scaling of trapped ions qubits is currently sourced from Analog Devices. Commercial silicon based photon only at the blueprint stage. counters would be used to read the quantum states with an efficiency of 30% could be sourced from Hamamatsu. Each microfabri- Compound semiconductor road map could be the wild card. We cated junction (with a qubit) would be manufactured on a traditional believe that the Bell Labs (owned by Nokia) have an aggressive road Silicon wafer substrate, with traditional etching and deposition tech- map for 2017 for delivering working gates (functions) around already niques - that said the prototype we saw in the lab was all hand made working qubits based on a very thin layer of gallium arsenide running so there is lots of potential for miniaturisation down to 1,296 junc- at very low temperature (below 1 Kelvin) - a recent Bell Labs paper tions per 4-inch wafer in a first stage. Aligning the modules and inter- in Nature and our discussion with CTO of Bell Labs Marcus Weldon connects in the order of 3 to 5nm is likely to be a technical challenge shows strong confidence in reaching the target to demonstrate the but, not surprisingly, ASML is suggested as the supplier of equipment gates by the end of 2017 and to then scale the qubits with a manufac- to achieve this. turing partner and an IP licensing programme for Nokia. The qubits last longer and thus have much lower error rates than with the other An experimental comparison of the two leading quantum computer materials according to Nokia Bell Labs. Compound semis is also architectures above (supraconducting junction and trapped ions) has researched by the University of Delft QuTech with a $50m invest- been established by a team of scientists including Christopher ment programme from Intel since 2015, and the University of Monroe of the University of Maryland in 2016. While the two archi- Purdue, with backing from Microsoft. NTT Basic Research Laborato- tectures have different ways of being programmed, the team has ries also has researchers working on this road map. established that the ion trapped architecture was slower but more precise. Also the trapped ion roadmap does not require sub 1 Kelvin Exhibit 13: Technology and company landscapes Source: Morgan Stanley Research 18
The core of the quantum computer thus already exists and is well Exhibit 15: understood, be it a Josephson junction or a trapped ion. The real ques- First integrated circuit (replica of the Texas Instruments Kilby proto- tion now is a how to scale and miniaturize those qubits. Referring type) back to the invention of the transistor effect in 1925 by Julius Edgar Lilienfeld, it took more than 20 years to implement it in real life in 1947 at the Bell Labs. The team lead by William Shockley won the Nobel Prize in Physics in 1956. In took another decade, for Jack Kilby at Texas Instruments to make the first integrated circuit chip, in 1958 (Kilby won the Nobel Prize in Physics in 2000). But this is a team of engineers moving from Fairchild to Intel, which created the largest semiconductor company of the past century. Hence, it is interesting to note that this is the company that industrialized the process that created the most value, not the one that invented the concept. Exhibit 14: First transistor designed at the Bell Labs in 1948 Source: Wikicommons While this is a team of scientists/physicists in one lab, larger companies like IBM, Intel, Google and Microsoft are also looking at the quantum computing opportunity. IBM and quantum computing - first commercially available small scale universal quantum computer We met with Dr. Arvind Krishna, Senior Vice President and director of IBM Research. IBM is dedicating considerable resources to the area of quantum computing. With a team of 100s of engineers, a long history in microelectronics, systems, and software, and strong relationships with large enterprises and governments, IBM is well positioned to lead in commercializing quantum computing. The company views the $5-6 billion high-end computing market as the initial target but sees the potential for new use cases to grow the addressable market for quantum computing to $10 billion-plus. Source: Wikicommons Morgan Stanley Research 19
In 2016, IBM launched Quantum Experience (now called IBM Q Expe- In particular, IBM explained that a laptop can simulate a molecule rience) with a 5 qubit universal quantum processor that is freely with 25 electrons but no classical computer could ever be built to accessible on the IBM Cloud, which was upgraded to a 16 qubit proc- simulate a 50-electron system as the problem is exponential by essor in 2017. IBM also developed a 17 qubit commercial processor nature. A quantum computer with 50 qubits could simulate that mol- prototype which will be the basis for the first IBM Q early-access ecule easily and IBM has already demonstrated the simulation of H2, commercial systems. IBM Q is an industry-first initiative to build com- LiH, and BeH2 molecules with 2, 4, and 6 qubits, respectively. mercially available universal quantum computers and services for business and science which will be delivered via the IBM Cloud plat- Supporting quantum software development is key. IBM provides form. The company is on the path to achieve a 50 qubit system within standard software wording (APIs) such that anyone can create the next few years which it believes represents the point quantum quantum algorithms and call them from a standard language computing can solve a wider range of complex problems far better (Python), hence reducing the time for a software engineer to retrain than a classical computer. Already, 50,000 users executed more to quantum computing. IBM is also supporting early customers. IBM than half a million trials on IBM Q Experience. Significant interest in is working with Raytheon to demonstrate the first proven examples the scientific community supports the view that the technology com- of quantum computers' advantage over a classical computer. With munity will embrace the platform as it scales. only a 5 supercomputing qubits process, the quantum algorithm consistently identified an unknown string of bits in a black box in 100- Already, use cases are emerging. At IBM's investor briefing in March fold fewer steps and was more tolerant to noise than the classical 2017, management explained that quantum computing will "enable a binary computing architecture. In layman's terms, the classical binary range of computation which would take more than the age of the computer was trying every possible combination, one by one, in a universe to do on classical computers," including for medicine and very orderly but slow manner, while the quantum computer quickly materials, machine learning, and searching big data. According to converged toward the right solution as if it was guessing the answer. IBM, a quantum computer with 10-30 qubits could beat any classical This is a strong step towards artificial intelligence as the human brain computer for molecule simulation (chemistry, pharma), and 1,000 would also start with an informed guess. qubits could supersede GPU-based inference calculations for artificial intelligence. Exhibit 16: Why quantum computing matters, according to IBM Source: IBM 20
Cloud delivery is the future of quantum computing. In 2017, IBM Exhibit 18: announced the IBM Q initiative to build commercially available uni- IBM 5 qubits chip versal quantum computing systems delivered via the IBM Cloud Platform. Following Watson and blockchain, the company believes quantum computing could form the next set of services delivered via IBM Cloud. Beyond the typical cloud advantages of cost and agility, quantum computing delivered via the cloud addresses a number of bottlenecks, including the need to operate in a very low temperature which requires specialized technology and resource constrained liquid helium. Exhibit 17: How to reach close to absolute zero temperature to slow down molecules and take control of quantum properties of molecules Source: IBM Exhibit 19: IBM 16 qubits chip Source: IBM Source: Blufors Cryogenics Morgan Stanley Research 21
Intel and quantum computing - partnership with Delft Univer- Google and quantum computing - the Quantum Ai Lab and collabo- sity and report to the US President ration with UCSB. We spoke with Jim Clarke, who heads Intel's quantum computing Google Research has a Quantum Ai Lab and has also worked with the efforts. D-Wave computer (not a universal quantum computer). Google has not communicated to the public for more than a year on its quantum Intel CEO Brian Krzanich is very excited about quantum computing computing research but given the potential cost savings on sorting and, like us, believes "it has the potential to augment the capabilities and searching algorithms we believe this is critical for Google's long- of tomorrow's high performance computers." Intel's CEO recognises term position and gross margin. the challenges including how to fabricate, connect and control many more qubits According to a report by Futurism.com, Google has improved its current quantum computer from 9 qubits (last official publication in When US semiconductor companies wrote to the US President at the 2015) to a 20 qubit chip currently under test and working to deliver beginning of 2017, they highlighted research in quantum computing a 50 qubit chip by the end of 2017. The last publication regarding the as one of the ways to leapfrog China's $100bn plan to build a semi- 9 qubit chip was particularly interesting regarding the reduction of conductor industry of its own (Semiconductors: Report to the Presi- errors in the qubits vs other companies, down to 99.5%. Should this dent Ensuring Long Term US leadership in Semiconductors ). The quantum computer based on 50 qubit have customised hardware quantum computer would develop "a pilot version of a high scale, then it is reasonable to believe it could be dedicated to number zero carbon, cost-competitive energy system," which we believe sorting and searching functions, thus at the core of Google business refers to the nuclear fusion system that many countries have been and that quantum computing could very well stay internal to Google. working on for several decades but failed to achieve because of the lack of computing power with the current computer architecture. Microsoft Looking to Build the Full Quantum Stack Intel launched its quantum computing program in 2015 in partner- Microsoft has been investing in quantum computing research for ship with Netherlands' Delft University of Technology. The company over 10 years. The appointment of Todd Holmdahl to head Micro- is hedging its bets by working on two different quantum computing soft's efforts in quantum computing in November 2016 marked the approaches simultaneously: 1) superconducting qubits (similar to transition of this research effort, into the realm of engineering and efforts by other companies in this space); and 2) silicon spin qubit development. In Todd's words, Microsoft saw the research, "at an (unique to Intel but something that is behind the progress made on inflection point, in which we are now ready to go from research to the superconducting qubit front). Having said that, the company has engineering" with a clear road map to a scalable quantum computer. little to publically show for its efforts to date with its silicon-based A twenty year Microsoft veteran, Todd had previously led efforts to efforts around only 2-3 qubits currently. The company expects to bring the Xbox, Kinect, and Hololens to market. The game plan at Mic- progress to 50 qubits in two years and 1000 qubits in 5-6 years, while rosoft is to develop the entire stack for the scalable quantum com- targeting commercial quantum applications in a decade from now. In puter, starting from the people pioneering the science, to addition, the company is focussed on developing complete quantum manufacturing the quantum hardware, to developing the software system that would include qubit chips and custom control elec- language with which new algorithms and program will be written. All tronics, while data processing could be boosted by high end FPGA/ these efforts are occurring in parallel, with a software language custom core. already in place (LiquiI>), the first qubit of Microsoft's topological quantum computer coming online imminently, and an existing service delivery model in Azure the most likely route to market. Our estimate is a commercial quantum computing service (a system of ~100 logical qubits) available from Microsoft within 5 years. 22
The People quantum state means each physical qubit represents a logical qubit, Through direct hires and collaboration, Microsoft has brought or said another way, competing systems need an order of magnitude together the following quantum computing researchers: more qubits to do the same amount of work as Microsoft's qubits. l Michael Freedman Head of Station Q Santa Barbara: A Microsoft is currently pursuing three parallel development paths for renowned mathematician and Fields Medal winner, manufacturing the physical quantum computer. These include semi- Freedman heads up a Microsoft Research lab located at the conducting nanowires, carbon nanotubes, and a CMOS based system University of Santa Barbara. Freedman has pioneered much (possibly with a superconducting layer on top of a compound semi- of the fundamental work of topological quantum computing, conductor substrate). The first qubits are expected to be online a differentiated approach to leveraging quantum effects for 'imminently', with a roadmap to 10's and 10,000's qubits well plotted computing. ahead. l Leo Kouwenhoven - Technical University of Delft: A Professor of Physics at TU Delft, Kouwenhoven was the founding The Software Already Coding Algorithms to Harness Quantum director of QuTech, an Advanced Research Center on Abilities Quantum Technologies, and leads research efforts around In line with the hardware efforts at Microsoft's Quantum Architec- potential quantum structures like semiconducting nanowires tures and Computation Group (QuArc), is the development of a soft- and carbon nanotubes. ware platform for quantum computing called LIQUil> (pronounced l Charles Marcus - Niels Bohr Institute: The Villum Kann Ras- liquid). LIQUil> includes a programming language, optimization and mussen Professor at NBI and a Center Director at the Center scheduling algorithms, quantum simulators and compilers for trans- for Quantum Devices, Charles Marcus' research interests lating high-level programs into low-level machine instructions for include fractional Hall systems (which give rise to topological quantum computers. The LIQUil> platform enables researchers and qubits). students to simulate quantum circuits, and use recently developed l Matthias Troyer - Professor of Computational Physics at ETH quantum algorithms to begin writing programs utilizing quantum Zurich: A professor at ETH Zurich, Troyer develops simula- computing power even before the actual quantum computers are tion algorithms for quantum systems. available. l David Reilly - Experimental Physicist at the Center for Quantum Machines at the University of Sydney: Professor D-Wave has been selling quantum computers since 2010 to part- Reilly leads research efforts on enabling technologies to con- ners including Lockheed Martin, Nasa and Google. Whilst D-Wave is trol condensed matter systems at the quantum level. not a universal quantum computer, i.e. it does not work with every workload/algorithm, its quantum computer has the largest claimed The Technology Pursuing Higher Fidelity with Topological number of qubit (2,048) and is best suited for the following tasks Qubits according the company: Optimization, Machine learning, Sampling / More traditional quantum computer architectures look to store and Monte Carlo, Pattern recognition and anomaly detection, Cyber manipulate information by capturing a particle in a localized space, security, Image analysis, Financial analysis, Software / hardware veri- for example capturing an electron to measure its spin. These systems fication and validation, Bioinformatics / cancer research. While the are very sensitive to the external environment, which introduces D-Wave quantum computer is not universal, D-Wave is the only com- errors into the system (causes decoherence of the qubit). Microsoft's pany to have sold quantum computers for several years to reputable quantum computing efforts base around the idea of a topological customers at a cost of over $10m per machine depending on the con- quantum computer, where the information is contained in an inher- figuration. Like IBM, the D-Wave computer is based on a supracon- ently more stable topological space the interaction of field lines ducting chip made of Niobium and there is a roadmap to make it with from multiple particles called Marojana fermions into what physi- Aluminum, according to an industry specialist who has been using the cists refer to as braids. Todd Holmdahl sees the stronger inherent D-Wave computer for many years in a US University. Discussion with fidelity of the topological architecture as the key to vaulting Micro- industry specialists show, however, that inputting data into the com- soft's efforts ahead of competitors. Thought of simply, traditional puter has proved difficult and the results were very noisy, i.e. with a quantum computers necessitate multiple physical qubits in a system high error rate to error correct against one logical qubit, which actually does the computational work. The inherent stability of the topological Morgan Stanley Research 23
Rigetti, a private company with $69m in funding formed by Chad Exhibit 20: Rigetti, who previously worked at IBM Quantum Computing Atos quantum computer simulator using classical computer architec- Group, announced in June 2017 an 8 qubit computer based on a ture supraconductor chip (aluminum) and a software environment linked to that chip called Forest. Rigetti has also launched a simulation platform (36 qubits with accurate noise simulation) running on classical computer chips, thus competing with Atos and Fujitsu. In terms of roadmap, Rigetti believes it can scale its quantum computer hardware to above 50 qubits in 2018. Alibaba In 2015, AliCloud/Aliyun announced a breakthrough in artificial intelligence and set up a Quantum Computing Laboratory with the Chinese Academy of Sciences (we have not soken with them). Atos - Paris-based Quantum research lab Atos has launched a quantum specific emulator based on classical computer architecture (CPUs, GPUs, FPGAs and ASICs) able to simulate 30 to 40 qubits at an affordable cost ($100k to $1m). The computer can be attached to a real physical quantum computer to offload some calculations. We believe this is a good approach for companies and universities willing to look at the opportunity to develop new algorithms and customise the attached quantum computing hardware. 24 Source: Atos
What is the potential impact of quantum computing to the stock market and the economy? Each industrial revolution has created large new sectors by market computer used to decode messages during WWII, there was only one cap. The first industrial revolution created the financial and transpor- and it was kept secret for a long time. tation sectors. The second one created the oil & gas, chemicals, pharma, and autos sectors. The third one created the tech and media At this stage, we view two main applications for quantum computers sectors, with new companies such as Amazon, Apple, IBM, Intel, Mic- - a useful one and a less so useful one. rosoft, Google, Samsung Electronics. If history repeats itself, then the fourth Industrial Revolution would create new companies and A quantum computer could be used to solve physical problems that new sectors. humans and supercomputers struggle to solve today, because they are exponentially complex by nature. We view BASF's aim to create That said, current large companies such as Google, IBM, Intel, and bespoke polymers as a good example, but it could be used to simulate Microsoft have already positioned themselves into quantum com- nuclear reactions like nuclear fusion reactors, which have proved dif- puting, but that does not mean they will succeed. There are also a few ficult to design in the past several decades (Super Phoenix and Iter private companies, with different levels of funding like D-Wave or reactors). It could also be used to design new molecules in the Professor Monroe's IonQ start-up, working on different types of Pharma sector. In summary, it could be beneficial to any company quantum computing as well as IdQuantique working on hack-proof looking for more affordable computing power today. transmission of data thanks for quantum physics principles. A quantum computer could be used to decode most types of cryptogtechnological disruptions are both risks and opportunities for sec- raphy keys used today. Current computers can multiply several tors and companies. The financial sector continued to thrive in the prime numbers easily, say 3*7*23*67=32,361 - even a pocket calcu- second and third/digital revolution, the autos, oil&gas and chemicals lator could do that. But a computer would take much more time to sectors continued to thrive in the digital revolution. But the train find the prime numbers making a large number like 32,361. Obviously companies / transportation sector did not do so well as the combus- most cryptography system would use many more prime numbers to tion engine gained share during the second industrial revolution. make finding the prime numbers very difficult. Your credit card secu- Even during the digital revolution, Intel and to a lesser extent Micro- rity, the sim card in your smartphone, the cryptography in your com- soft did very well until the launch of Windows 95 but have underper- puter are all based on this. A quantum computer would crack the formed since. code more easily and there could in theory be a transition period where cybersecurity/payment companies could become irrelevant In all scenarios, we believe that quantum computing will not super- unless they find quantum proof cryptography systems. sede all current forms of computing. Like when combustion engines started to gain traction, steam engines were still used for several dec- That does not look like very many use cases, but this is only the begin- ades for manufacturing applications and mass transportation. Based ning and use cases might increase as research into the field continues on the prototypes we have seen, using lasers, we believe that to progress. Who would have thought that the computer used to quantum computing will not be miniaturized to the point that it can decode messages during WWII or the microcontroller used to guide be used in any portable applications, i.e. a laptop, tablet, smartphone. the Apollo rocket to the moon could be used to send messages in Nor does it make much sense to use it for a basic server and to store your pocket, turn on and off the lights at home, and measure the dis- data. But there could be a handful very large quantum computers by tance to the car in front etc. the end of the decade, available to share, in the cloud, or not, held by private companies for their own use or large States. Like the very first Morgan Stanley Research 25
Could, for instance, quantum computing be used for Artificial Intelli- Machine learning requires a significant amount of processing power gence applications? NASA, Universities Space Research Association and thus quantum, at least theoretically, should be a prime candidate and Google Research set up the Quantum AI Lab in 2013 and pub- to take AI beyond the current state-of-the-art calculation of infer- lished reports comparing their efforts based on a D-Wave quantum ences, as shown in the most recent demonstration made by IBM and annealing computer in 2014. That said, reports from the lab have Raytheon that a quantum computer would guess an unknow string been scarce since 2015. nvidia has stolen the show with so much of bits in a black box in 100 times fewer steps than a classical binary progress made in AI thanks to AI specific chip architecture and an easy computer using "if, then, else" loops. to learn language (Kuda). Google has a Quantum AI Lab and IBM's Director of Research Dr. Arvind Krishna has stated publically at its There are different paths of development shown in the following 2017 Investor Briefing 2017 that quantum computers could be used quadrant chart. On the x-axis, the speed of development from here, for Machine Learning (and Big Data searching). fast or slow. On the y-axis, if the technology will remain public or be widely available as a cloud service Exhibit 21: Potential outcomes of the quantum computing depending on take up and on business model (cloud vs internal) Source: Morgan Stanley Research estimates 26