SystemX

Buildings account for the largest share of U.S. greenhouse gas emissions at 32% of the total, followed by industry at 30%, transportation at 29%, and agriculture at 9%. Clearly, any program to greatly reduce greenhouse gas emissions must include buildings. This talk will explore the opportunities and challenges to greatly reducing greenhouse gas emissions in the building sector, using as a case study a recent solar retrofit of an existing house that rendered it "zero-carbon." Implications for the future of the utility system will be discussed.

Google's Tensor Processing Unit (TPU), first deployed in 2015, provides services today for more than one billion people and provides more than an order of magnitude improvement in performance and performance/W compared to contemporary platforms. Inspired by the success of the first TPU for neural network inference, Google developed multiple generations of machine learning supercomputers for neural network training that allow near linear scaling of ML workloads running on TPUv2 and TPUv3 processors. TPUs extend research frontiers and benefit a growing number of Google services.

By the end of 1990 the Acorn RISC Machine, which was designed in Britain, was practically extinct. The parent company, Acorn, was down a very dark financial alley, and the remnants of the design team were cast out to fend for themselves, equipped with about 18 months of financial rations from Apple, and a really rather odd microprocessor design. When Dave Jaggar joined ARM a few months later, with the ink not quite dry on his Master's Thesis, he thought perhaps he'd made a dreadful mistake. However, after twelve months he was given free range to start the ARM architecture afresh, and the new Advanced RISC Machine, as the company was now named, was born. Over the next 8 years he worked out a little bit about computer architecture, in the same way that a 17th century surgeon understands anatomy, then he was made the first ARM Fellow, so he promptly retired back to New Zealand to raise his children. As it's the 25th anniversary of his quite successful Thumb compressed instruction set, and his children have all left home, he's been told it's about time he explained himself.

The current electronics industry has been completely dominated by Si-based devices due to its exceptionally low materials cost. However, demand for non-Si electronics is becoming substantially high because current/next generation electronics requires novel functionalities that can never be achieved by Si-based materials. Unfortunately, the extremely high cost of non-Si semiconductor materials prohibits the progress in this field. Recently our team has invented a new crystalline growth concept, termed as "remote epitaxy", which can copy/paste crystalline information of the wafer remotely through graphene, thus generating single-crystalline films on graphene [1,2]. These single-crystalline films are easily released from the slippery graphene surface and the graphene-coated substrates can be infinitely reused to generate single-crystalline films. Thus, the remote epitaxy technique can cost-efficiently produce freestanding single-crystalline films. This allows unprecedented functionality of flexible device functionality required for current ubiquitous electronics. In addition, we have recently demonstrated a manufacturing method to manipulate wafer-scale 2D materials with atomic precision to form monolayer-by-monolayer stacks of wafer-scale 2D material heterostructures [3]. In this talk, I will discuss the implication of this new technology for revolutionary design of next generation electronic/photonic devices with combination of 3D/2D materials.

Lastly, I will discuss about an ultimate alternative computing solution that does not follow the conventional von Neuman method. As Moore's law approaches its physical limits, brain-inspired neuromorphic computing has recently emerged as a promising alternative because of its compatibility with AI. In the neuromorphic computing system, resistive random access memory (RRAM) can be used as an artificial synapse for weight elements in neural network algorithms. RRAM typically utilizes a defective amorphous solid as a switching medium. However, due to the random nature of amorphous phase, it has been challenging to precisely control weights in artificial synapses, thus resulting in poor learning accuracy. Our team recently demonstrated single-crystalline-based artificial synapses that show precise control of synaptic weights, promising superior online learning accuracy of 95.1% – a key step paving the way towards post von Neumann computing [4]. I will discuss about how we design the materials and devices for this new neuromorphic hardware.

Further scaling of complementary metal-oxide-semiconductor (CMOS) dimensions will soon lead to a tremendous rise in power consumption while limited gain in the performance of integrated circuits. "Beyond-CMOS" devices, based on new materials, device concepts and architectures, can potentially overcome these limitations and further improve the performance, reduce energy consumption, and add novel functionalities to the CMOS platform. In this talk, I will present nanoscale electronic and photonic devices based on two-dimensional (2D) materials and ferroelectric materials. In particular, I will discuss the logic devices, RF devices, photodetectors, plasmonic devices, and tunneling devices based on graphene and transition metal dichalcogenides. I will also present our recent results on non-volatile memories and ferroelectric tunneling junctions (FTJs) based on ferroelectric hafnium oxide and 2D ferroelectric indium selenides.

The field of AI was motivated originally by the objective of automating tasks performed by humans. While advances in machine learning have enabled impressive capabilities such as self-driving vehicles, more cognitive tasks such as planning and design have resisted full automation because of the vast amounts of knowledge and commonsense reasoning that they require. This talk describes a line of research aimed at developing AI systems that are designed to augment rather than replace human capabilities, leveraging automated planning, machine learning, and natural language understanding technologies. It also presents several successful applications of the research in deployed systems.

From the beginning of recording history, mapping the world has been critical to human growth and exploration. Today, mapping the world in 3D is equally critical to the future of AR and how we all will experience the world around us.

To usher Solar 2.0 beyond incremental solar module cost and efficiency improvement, it is critical to understand true metrics for solar projects that impact the ultimate cost of electricity. This talk is divided in three parts. The first part discusses the system's and project relevant metrics which constitute the bottom line for customers. It also sets the stage for the importance of a multi-dimensional advancement of solar module technology as apposed to conventional incremental improvements in cost and efficiency areas only. The second part discusses Solar 1.0, the historical drivers which contributed to the dramatic cost reduction and the rapid commoditization of solar PV over the last decade. The final part of the talk delves into potential disruptive ideas both related to dramatic efficiency improvement and system's level smartness, which will drive the next wave in advancement, the Solar 2.0.

Quantum computing holds the promise to transform many industries, by rendering some of todays intractable problems feasible. To do so, quantum computers are built upon fundamentally different rules from standard computers, harnessing the bizarre but beautiful quantum mechanical laws that underpin the behavior of atoms and molecules. This will transform cryptography, material and chemical design, among other key areas. I will attempt to demystify some key aspects of quantum computing, highlight some of the key industrial applications, and will outline Microsoft's full-stack approach to overcome the challenges to build scalable quantum computers.

Quantum computing holds the promise to transform many industries, by rendering some of todays intractable problems feasible. To do so, quantum computers are built upon fundamentally different rules from standard computers, harnessing the bizarre but beautiful quantum mechanical laws that underpin the behavior of atoms and molecules. This will transform cryptography, material and chemical design, among other key areas. I will attempt to demystify some key aspects of quantum computing, highlight some of the key industrial applications, and will outline Microsoft's full-stack approach to overcome the challenges to build scalable quantum computers.