November_EDFA_Digital

edfas.org 51 ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 22 NO. 4 see the quantumcomputer equivalent of aworkstation or personal computer inside of 25 years. Why go through all this effort? The short answer is deceptively simple; early research suggests that quantum computing might accelerate deep learning (DL) and artificial intelligence (AI) computational capabilities by orders of magnitude over planned semiconductor-based architectures. To do so, quantum computers will need thousands of error-corrected qubitsmaintaining simulta- neous coherence for long enough to train the equivalent of today’s complex ensembles of DL models. But that is not enough. Data scientists will also need access to quantum DL software development frameworks that do not require working knowledge of quantum physics. That challenge is being addressed by a few quantum computing software framework startups. However, those startups will eventually need commercial hardware to run on. In 2018, the largest announced general-purpose quantum computer architectures reached the 49-72 non-error-corrected qubit range. They have not publicly progressed since then, including Honeywell’s quantum computer announcement earlier this year. It is likely that 15 years is an optimistic estimate for commercially viable quantum cloud computing, despite the marketing enthusiasm of many early quantum com- puter systems vendors. Conversely, there are good oppor- tunities in metrology and components development in the meanwhile. Commercializing the non-quantumpieces of quantum computing will enable quantum computer developers to focus their research time and money on advancing archi- tectural research. ABOUT THE AUTHOR Paul Teich is Liftr Insights’ resi- dent expert for tracked services and technologies. He is fluent in com- puter architecture, artificial intelli- gence, edge computing, and Internet of things. Prior to Liftr Insights, Teich was a principal analyst at TIRIAS Research, a senior analyst for Moor Insights & Strategy, and worked for AMD for two decades. Paul holds a B.S. in computer science from Texas A&M and an M.S. in tech- nology commercialization from the University of Texas’ McCombs School of Business. He received 12 U.S. patents while working in marketing and strategy roles at AMD. to maintain coherence and interact with each otherwithout losing their quantum states due to electromagnetic interfer- encewithin the quantum computer itself. Error detection and correction : Quantum states are finicky. But then so are 7 nm FinFET gates. To move past analog quantum com- puting using only dozens of qubits, quantum error correction codes must be developed. And for that to happen, quantum computer architectsmust detect when errors occur. For an idea of how basic current quantum computing is, recent ex- periments have: • Increased spin-orbit qubit coherence times several orders of magnitude to ~10 milliseconds • Raised the opera- tional temperature of individual qubits two orders of magnitude to 1.5 Kelvin Inaddition to the above, there are no standards for and few common approaches to general-purpose qubit architecture. Every quantum computer development team is designing a bespoke architecture usingdifferent approaches todefine, capture, and control qubits. Each bespoke architecture therefore presents different challenges for isolating andmanipulat- ing qubits and then detecting the results of calculations. Because of the experimental nature of quantum com- puters and their environmental requirements, they are large and require constant attention and maintenance. They should properly be called “quantum cloud comput- ers,” as most commercialization plans (outside of a few government labs and Global 100 purchases) will make quantumcomputers cloud accessible. It is unlikelywewill IBM Q quantum computer thermal interface, 2018. (Source: Paul Teich) Each layer of this quantum computer chandelier spans a thermal gradient in a diffusion refrigerator, starting at ~40 K at the top, then 4 K, etc., with the qubit cannister at the bottom operating at 15 mK. Microwave conduit is used instead of electrical wiring to transmit signals across the extreme thermal gradients.