key: cord-0555620-6se1pii7 authors: Wu, Carole-Jean; Manne, Srilatha; Ranganathan, Parthasarathy; Bird, Sarah; Greenstein, Shane title: Socio-Technological Challenges and Opportunities: Paths Forward date: 2021-08-15 journal: nan DOI: nan sha: 33129ee647593b873fed4477a056c750a78c2684 doc_id: 555620 cord_uid: 6se1pii7 Advancements in digital technologies have a bootstrapping effect. The past fifty years of technological innovations from the computer architecture community have brought innovations and orders-of-magnitude efficiency improvements that engender use cases that were not previously possible -- stimulating novel application domains and increasing uses and deployments at an ever-faster pace. Consequently, computing technologies have fueled significant economic growth, creating education opportunities, enabling access to a wider and more diverse spectrum of information, and, at the same time, connecting people of differing needs in the world together. Technology must be offered that is inclusive of the world's physical, cultural, and economic diversity, and which is manufactured, used, and recycled with environmental sustainability at the forefront. For the next decades to come, we envision significant cross-disciplinary efforts to build a circular development cycle by placing pervasive connectivity, sustainability, and demographic inclusion at the design forefront in order to sustain and expand the benefits of a technologically rich society. We hope this work will inspire our computing community to take broader and more holistic approaches when developing technological solutions to serve people from different parts of the world. Moore's law scaling underpins the evolution of microprocessors [Moore, 1965] . The steady doubling of transistor density enables miniaturization of computing systems, from large mainframes to personal computers and from mobile/smartphones to Internet of Things (IoTs) and AR/VR wearables. The 1990s were the golden age of microprocessor innovations. Microarchitectural optimizations enabled impressive ILP scaling: most notably, in-order vs. out-of-order execution [Tomasulo, 1967 , Smith, 1982a , Hwu and Patt, 1986 , branch predictors [Smith, 1981 , Lee and Smith, 1984 , Pnevmatikatos et al., 1993 , Yeh et al., 1993 , Jimenez and Lin, 2001 , caches [Smith, 1982b , Hill, 1988 , Przybylski et al., 1989 , Jouppi, 1990 , prefetchers [Baer and Chen, 1991 , Fu et al., 1992 , Falsafi and Wenisch, 2014 , single vs. simultaneous multithreading [Tullsen et al., 1995 , Nemirovsky and Tullsen, 2013 . [Google, a] whereas the PUE of Facebook datacenters is 1.10 (2020) [Facebook] and the average PUE for a typical data center in 2020 is 1.58 [Lawrence, 2019 [Lawrence, , 2020 . In the 2000s, microprocessors faced two significant challenges: the memory wall [Wulf and McKee, 1996] and the power wall [Dennard et al., 1974 , Bohr, 2007 . While processor frequencies improved with Moore's law scaling, memory latency did not and memory subsystems increasingly gated performance. Furthermore, Dennard scaling came to an end and fine-grained, high power density thermal hot spots limited the performance of microprocessors. The Memory and Power walls subsequently drove decades of innovations in multi-core scaling [Olukotun et al., 2007] , memory consistency and cache coherence [Adve and Gharachorloo, 1996 , Hill, 1998 , Sorin et al., 2011 , cache and memory hierarchy optimization [Qureshi et al., 2007 , Jaleel et al., 2010 , Wu et al., 2011 , Balasubramonian et al., 2011 , Sardashti et al., 2015 , Balasubramonian, 2019 , Jain and Lin, 2019 , network-on-chip design and optimization [Dally and Towles, 2001 , Wang et al., 2002 , Enright and Peh, 2009 , Jerger et al., 2017 , and power-and thermal-aware design and management [Brooks et al., 2000 , Brooks and Martonosi, 2001 , Skadron et al., 2003 , Kaxiras and Martonosi, 2008 , Själander et al., 2014 . During the same period, computations were migrating from client/personal devices to the cloud, demanding significant investment in large-scale data centers [Barroso and Hölzle, 2009] . The location of these data centers is dictated by a myriad of constraints -maximizing power and operational efficiency, proximity to population centers, weather conditions, local tax breaks -leading to some interesting tradeoffs in size, location, and ownership (on premise or cloud based) of the data centers. In 2019, urban data centers that optimize for service latency responsiveness are 26.7% smaller in size than the average data center operated by the major cloud providers [Greenstein and Fang, 2020] . Furthermore, between traditional and highly optimized hyperscale data centers, power usage effectiveness (PUE) has a stark difference -more than 40% higher efficiency for hyperscale data centers (Figure 1 ). Going forward, the demand on fast(er) service response and availability is playing an increasingly significant role on data center site selection and computing infrastructures connecting the edge and the cloud. The 2010s is the golden age of domain-specific architectures and specialized hardware, fueled by the rise of big data and AI [Esmaeilzadeh et al., 2011 , Jouppi et al., 2018 . The massive economic growth opportunities of AI have revolutionized the entire system stack design, resulting in hardware tailored to machine learning execution [Ovtcharov et al., 2015 , Chen et al., 2016 , Jouppi et al., 2017 , Fowers et al., 2018 , Shao et al., 2019 , Mattson et al., 2020 , Henry et al., 2020 , Reddi et al., 2021 , Anderson et al., 2021 , Jang et al., 2021 , Thompto et al., 2021 from megawatt data center-scale infrastructures, to tens of watts inference engines, to micro-watt microcontrollers at the edge. Building on top of domain-specific characteristics, further system efficiency can be extracted with a rich array of application-specific accelerators at the cloud scale [Taylor et al., 2020 , Ranganathan et al., 2021 . The last fifty years of digital products have been driven by a combination of market innovation and user needs. In some cases, scientific curiosity, coupled with a market need for specialty medical drugs drove innovations such as the pursuit of the human genome. In other cases, innovations created a market such as the case for smartphones. The availability of smartphones led to further innovations and massive disruptions via sharing economy companies such as Uber and AirBnB [Hamari et al., 2016] . None of this would have been feasible without fundamental innovation in process technology, hardware and software design, and a laser focus on efficiency (Figure 1 ), optimization (Figure 2) , and cost reduction. Altogether, digital technology advancement has led to efficiencies of scale propelling decades of economic growth. The narrow focus on marketability, efficiency, and disruptive innovation has also resulted in many challenging, and sometimes unexpected, societal issues. Examples include widening disparity and inequity of access to digital technologies, spread of disinformation and misinformation in online platforms, privacy and security violations at both the personal and political level, and propagation of human bias into AI training and use cases at-scale [Simons and Jones, 2012 , Abelson et al., 2015 , Gervais et al., 2016 , Ahmed et al., 2017 , Whittaker et al., 2018 , Speicher et al., 2018 , Chouldechova et al., 2018 , Ekstrand et al., 2018 , Selbst et al., 2019 , Ali et al., 2019 , Babaei et al., 2019 , Crawford et al., 2019 , Papakyriakopoulos et al., 2020 , Bender et al., 2021 , Kleinberg and Raghavan, 2021 . In addition, the information and computing technology sectors consume a significant amount of global electricity, water, and natural resources, leading to paramount carbon and environmental footprint [Jain and Wullert, 2002 , Strubell et al., 2019 , Manne, 2020 , Crawford et al., 2021 . Digital technology is an integral part of human existence. As the field matures and as the world faces dire challenges from climate change to societal and political upheavals, a deliberate approach is required to technological innovation that centers on a positive societal impact while serving the needs of a market economy. Predicting the future is always difficult. How many of us could have predicted that the technology depicted in futuristic TV shows from five decades ago such as Star Trek would be commonplace today (cell phones, AI, voice recognition to name a few)? However, what is possible to foretell are the following: • Pervasive Connectivity: the internet and the technology enabling it will become even more fundamental to everyday life, requiring that information and communication technology infrastructure, just like power grids, be both secure and resilient; • Sustainability: Environmental pollution, resource depletion and climate change will dominate how products are designed, produced, consumed, and recycled. Systems must be manufactured with less planetary impact, use less energy while in operation, and produce less e-waste at the end of life; and • Demographic Inclusion: Massive demographic upheavals resulting in an aging population in most of the world and rising population centers in Sub-Saharan Africa will influence what products are created and for whom, and where technology is designed and deployed in the future. Pervasive Connectivity. In 2021, approximately 65% of the world's population has access to the internet [Lin, 2021] , and this is expected to improve moving forward. Seamless connectivity will be required for everyday activities from ordering groceries to driving your car. There are advantages to connectivity from more efficient driving and less food waste, easy access to educational resources, and the ability to work remotely. However, what this also implies is that our lives and livelihoods will be inexorably linked to the accessibility, privacy, security, and resilience of the IT infrastructure. Reliable internet access for the world's population is an essential requirement of pervasive connectivity. Despite the positive societal and economical impact, at-scale power delivery and networking infrastructure development has proven to be costly and highly geographically-constrained. According to an analysis from the FCC, at least 18 million Americans did not have stable broadband in 2020 [Wheeler, 2020] , and many who have access may not be able to afford it. According to a recent UN report, two-thirds of school age children do not have access to the internet in their home [unicef, 2020] . Diverse technology innovations are particularly needed to increase information accessibility and to connect people of differing needs in the world together in a resilient manner. Microsoft Airband [Microsoft, a] aims to expand broadband access in rural parts of the United States while Google Loon [Loon] and Facebook Aquila [Wikipedia] provide affordable high-speed internet to under-connected communities by overcoming physical barriers with innovative technology solutions. The recent success of placing internet communication satellites in low Earth orbit opens the door to tremendous opportunities for providing internet access to populations in challenging geographic locations [Starlink] . While satellite communication removes the geographical and political boundaries, it can accelerate the spread of (mis)information. Thus, as technologies continue to enable pervasive connectivity at scale, we must take a deliberate approach to develop technology solutions responsibly. Along with pervasive connectivity comes the requirements for security and privacy given the prevalence of ransomware attacks and ongoing privacy hacks from both private groups and nation states. Many aspects of our public and private lives will be online and accessible to bad actors, and the internet and the technology it enables will continue to be susceptible without strong security and privacy measures in place from the top of the software stack to the underlying hardware running the code [Lipp et al., 2018 , Kocher et al., 2019 . The IT infrastructure, just like in other critical infrastructure such as telecommunication and power delivery, will require backup and recovery mechanisms in order to minimize or eliminate downtime. These already exist in cloud data centers with examples such as redundant storage [Microsoft, 2021a] and network [Microsoft, 2021b] , and backup batteries and generators. However, redundancy can require significant additional hardware and network bandwidth which inflates the cost of the overall cloud infrastructure. Recent climate events are pushing the issue of resiliency to the forefront. In the Texas power outages of 2021, many top-tier data centers were able to continue operations using diesel backup generators. However, there is a limit to how long backup systems can operate given the finite amount of fuel onsite and weather conditions making it difficult to transport more fuel to the data center [Sverdlik, 2021] . As catastrophic weather events become more common and last longer due to climate change, even the largest data centers may have trouble [Department of Energy, 2019]. For instance, the latest fires in Oregon brought down major power grid infrastructure that connects the grids of California to Oregon at the same time as a heat wave resulted in a need for additional power for cooling [Roth, 2021] . California narrowly escaped rolling blackouts this time, but these events will continue to increase in intensity and frequency as hot and dry conditions lengthen and strengthen the fire season, reduce hydro-electric power due to a lack of water, and increase the need for power for human comfort and survival. For IT infrastructures of the future, a one-size-fits-all solution may not work either for the type of data center or for the services that it provides. Large data centers that require many megawatts to operate may not be viable during a blackout with limited power and where critical operations such as hospitals are given priority. Smaller data centers may be more resilient either with backup generators or using a local micro-grid to run operations when the main grid is down. On the service side, some applications may be more tolerant of outages than others. Hence, with a limited power budget, application availability should be tiered during critical times. Other solutions include making applications resilient by design or more amenable to migration to a non-impacted data center given enough warning before catastrophic events occur [Hornsby, 2018, Google Cloud Architecture Center] . Designing for resilience addresses issues that are not just related to resilience under climate duress but also solutions that are tolerant of different geographies, power and networking infrastructures. In order to achieve pervasive connectivity across the world, IT infrastructure must be available across developed and developing regions. In Africa, for example, it is difficult and expensive to develop a power grid that can serve such a large continent. Renewables, however, have the benefit of being portable and scalable, and are powering much of Africa's latest power advances, with off-grid and micro-grid renewable energy solutions serving poorer and/or non-centralized communities [The World Bank, 2020]. Data centers serving these regions may also have to rely on micro-grids running on renewable energy that fluctuate in their capacity based on the weather. Hence, rather than designing for a resilient and consistent power grid, IT infrastructure may be better served by (co-)designing agile applications and data centers, such as [Íñigo Goiri et al., 2015 , Radovanovic et al., 2021 , that implicitly assume power fluctuations and variability. Achieving pervasive connectivity in the presence of resiliency, security, and privacy requirements can be challenging given the interrelated nature of the problems. For instance, current approaches to achieving resilient computing often rely on replication and redundancy which not only increases cost but also enlarges the surface of security and privacy preservation. Migrating applications to another region, which may be necessary due to power constraints, can expose data to further security threats. To retain physical security of data, applications may be hosted in limited geographical regions but this can increase application vulnerability to catastrophic events. In addition, secure and resilient computing infrastructures can often come with significant environmental implications. Hence, any computing infrastructure solutions must be cognizant of the multifaceted nature of the problems being addressed. Sustainability. Resource limitations, climate change, water depletion, electronic waste, ecosystem damage, and environmental racism are just a few of the topics under the larger sustainability umbrella. There is increased focus on these topics from both industrial and political institutions. All major technology companies have pledged to reduce or eliminate their carbon footprint in the next decade by reducing the environmental impact associated with manufacturing and using their products. Examples of such commitments are Facebook achieving NetZero in operational emissions in 2020 and across its value chain by 2030 [Facebook] , Apple's pledge for 100% carbon neutral supply chain by 2030 [Apple] , Microsoft's goal of being carbon negative by 2030 [Smith] , and Google's aim of 24x7 carbon free data centers [Google, b] . In 2020, Amazon, Google, Facebook, and Microsoft were the top four technology companies that purchased significant renewable energy capacities, accounting for 30% of the cumulative total from corporations globally [Schechner, 2021] . In addition, countries and trading zones are legislating carbon emission requirements. China has committed to be carbon free by 2060 [Myers, 2020] and the EU has committed to cut carbon emissions by 55% by 2030 [BBC, 2021] . Sustainability targets and the associated regulations will continue to increase, and hardware must be manufactured with less planetary impact, use less energy while in operation, and produce less e-waste at the end of life [Orcuttarchive, 2015 , Chang et al., 2017 . Existing practices such as the move to hyperscale data centers have already reduced IT's carbon footprint by consolidating and sharing computing resources, and operating those resources more efficiently (Figure 1 ) -AWS [Amazon, 2019] , Azure [Microsoft, 2020] , Google cloud [Evans and Gao, 2016] , and Facebook datacenter infrastructures [Lee and Rowe, 2020] . In fact, data center electricity consumption has slowed down significantly. The total energy consumption of the US data centers increased by about 4% from 2010-2014, compared with the estimated 24% increase from 2005-10 and nearly 90% increase from 2000-05 [Masanet et al., 2020] . Furthermore, despite the increase in the global data center energy use, the number of users benefiting from the cloud infrastructure have increased even more-from 2010-18, the global data center compute instances increased by 5.5 times with an estimated 6% increase in the global data center energy use. However, more can be done -we must go beyond efficiency optimization and build a sustainable ecosystem to achieve environmentally-sustainable computing. For instance, develop expandable hardware and software stack that facilitate significantly longer lifetimes than the current averages of less than 3 years for cell phones [Cordella et al., 2020] and 4 to 5 years for servers [Ascierto and Lawrence, 2020] . Modular system design will enable component-level upgrades without having to decommission the system at its entirety, reducing overall electronic waste and the environmental footprint [Fairphone] . Other actions being discussed or implemented require manufacturers to make their systems more repairable, resulting in increased product lifetime [Wiens and Gordon-Byrne, 2017, Alsever, 2021] . Another option is to reduce the number of devices in an average household. In the US, for example, the average household is equipped with an average of 25 connected devices [Deloitte, 2021] . In many cases, smartphones might be powerful enough for the task, but a tablet or laptop is needed for a viable keyboard or a larger viewing platform. Some of these additional devices could be replaced with a virtual reality or augmented reality solution that is both portable and can provide virtual keyboards and visual clarity without requiring additional hardware. To achieve an environmentally sustainable computing future, we will have to build a circular economy for computing that supports the design principle of reduce, reuse, repair, and recycle. These and other potential solutions likely require a complete redesign of the software and hardware stacks both at the edge, within the cloud, and in the edge-cloud collaborative execution environment, in order to provide resilient, long lasting, innovative solutions. Making technology more sustainable is only one part of the technical challenge. There is another side to the story-there are significant sustainability benefits resulting from computing technology. Programs, such as Farm Beats [Microsoft, 2015] , address how to optimize operations on a farm using technology that is inexpensive and readily available to people in rural areas. Food production accounts for 19% of the world's carbon emissions, and producing food more efficiently and with less toxicity has long term benefits for the world [Gates, 2021] . For the residential sector in the US, space heating and cooling contributed to over 40% of the total electricity consumption [U.S. Energy Information Adminstration]. This is where smart home IoT devices, such as Nest, can have an impact. AI is used to discover new electrocatalysts for more efficient and scalable ways to store and use renewable energy [Zitnick et al., 2020] while also being used to predict renewable energy availability ahead of actual generation to better utilize the energy [Elkin and Witherspoon] . Another example is the current Covid outbreak. As horrendous as the outbreak has been and continues to be, technology has enabled a portion of the economy to continue to operate even as employees work remotely. In addition, the global carbon emissions for 2020 dropped by 6.4% with vehicle transportation in the US accounting for a portion of the global reduction [Tollefson, 2021] . Looking forward, information technology can improve efficiencies in practically every sector, from manufacturing to food production to transportation to controlling the climate in our homes and offices. Although there is a carbon cost associated with manufacturing and operating the IT ecosystem, this cost must be evaluated holistically [Chang et al., 2010 , Bardon et al., 2020 , Patterson et al., 2021 in light of the benefits such an ecosystem can provide in other domains [Hager et al., 2019 , Tomasev et al., 2020 , Mulhern, 2021 , United Nation, 2021 . Demographic Inclusion. Recent data has pointed to significant demographic changes that will be occurring by the year 2100 [Vollset et al., 2020] . These include a radically declining and/or aging population in Europe, North America and Asia, and an increasing population of younger workers in Sub-Saharan Africa. Figure 3 shows one population projection using data from the United Nations [Roser, 2019] , indicating flattening or decreasing populations in most of the world except for Africa. The population decline is strongly correlated with an improved quality of life and educational achievement of women -much of which is being facilitated by technologically engendered developments such as portable devices and the internet -along with access to reproductive services. This trend holds regardless of the country, culture, or religion. These demographic changes will have a seismic impact on all societies, and will also dominate what and where technology is designed and deployed, respectively, in the future. For technology to be truly inclusive in a fully-connected world, it must be available and usable for everyone -regardless of physical capabilities, geographic restrictions, or economic constraints. With the projected demographic changes, assistive technologies to address the needs of an aging population with physical restrictions will be essential for enhancing an individual's ability to be a viable and contributing member of society. There is a rich history of intersection between assistive technology and everyday products. Word prediction was used as an assistive technology aid for individuals with physical or developmental difficulties before becoming more mainstream [Jacobs, 2015] . Close captioning helps the hearing impaired, but it is also used by people with average hearing to capture the conversation in a noisy environment or to help with accents in the dialogue. The reverse is also true -some of the most innovative technologies in the last few years are also inclusive of people with physical limitations. Smart speakers, such as Siri or Alexa, enable people with limited sight or physical mobility to communicate via verbal commands. Individuals with hearing issues can use their Apple AirPods as basic hearing aids [Haselton, 2019] , and people with average hearing can use AI-enabled voice assistants to eliminate background noise and focus on the conversation at hand [Nachmani et al., 2020] . For the sight impaired, virtual reality glasses are improving the eyesight of patients with degenerative eye disease [Fearn, 2020] . The virtual office technology and work culture developed during the Covid pandemic will continue to benefit individuals with physical limitations who may be unable to travel to work. Similarly, with the largest population of young workers residing in Africa over the next decades, innovative virtual office technology that can create seamless cohesion across geographies and cultures will be critical to economic success. Looking to the future, there is a clear demand for high-quality virtual reality experience in the first-person perspective. However, realizing smooth virtual space experience requires disruptive technology innovations -hardware architectures beyond Von Neumann, high pixel density displays [Ghosh et al., 2016 , Kubota, 2020 , near/in-sensor computing [Kodukula et al., 2021] , flexible electronics [MacKenzie and Ho, 2015] , or even brain-computer interface solutions [Brunner et al., 2011 , Karageorgos et al., 2020 . Another requirement of inclusive technology, especially in light of projected demographic changes, is that it must operate across different geographic boundaries. In 2021, much of modern technology assumes some degree of development and infrastructure, whether it be availability of the latest computing systems, broadband access, cell towers or power sources. However, much of the current world, including parts of developed countries, do not currently have unfettered access to these facilities. Even more daunting, approximately 770 million people, or about 1 out of 10 people in the world, do not have access to a stable supply of electricity [International Energy Agency] . In order to meet the upcoming needs of the world, technological growth and resources must focus on under-served areas. This includes developing data centers for regions that may not have a resilient power infrastructure or an infrastructure that is based on local renewable energy sources and does not tap into a regional grid. IT devices must be operable under adverse conditions where the local network and/or power may not be available 24/7 or varies based on external conditions [Isaacman and Martonosi, 2011 , Hester et al., 2013 , Lucia et al., 2017 . For instance, devices need to operate for days without a recharge or must operate locally until wireless or broadband is available [Yetim and Martonosi, 2015] to ensure delay tolerance while maintaining user experiences. Energy storage technology plays a crucial role to smooth the intermittent nature of renewable energy generation but the cost must be significantly improved for practical deployment [Plumer, 2021] . And, for technology to be truly inclusive, the way AI technologies are developed and used must be human-centered, driven by the cultural and demographic differences in the population and with pro-social goals [Stray, 2021] . Furthermore, given its increasingly large impact on the society, AI must be developed with deliberation to construct fairer and more inclusive decisions and, at the same time, it must be adopted responsibly [Askell et al., 2019 , NIST, 2021 . AI-powered products must be transparent with users on how data is gathered, used, and stored, and with controls to disable it [IBM, Microsoft, b, Google, c, Facebook, DOD] . Finally, any IT technology must be economically-accessible to most of the world's population. Unfortunately, even the most basic form of computing devices, a cell phone, is prohibitively expensive for much of the world. According to the Alliance for Affordable Internet, nearly 2.5 billion people live in countries where a basic cell phone would cost nearly a quarter or more of the average monthly income [Roser, 2020] . Making the internet available to everyone will not solve the problem if the devices commonly used to access the internet are a luxury item for the world's poor [Naseem et al., 2020] . In addition, many of the devices currently being used are inexpensive basic phones. Hence, solutions must be designed for "dumb" phones in order to reach the poorest communities. Even if users eventually have smartphones, data rates and lack of free wifi will limit the utility of data-heavy applications. Without reasonable device and application availability for the poorest communities, reaching the goal of pervasive connectivity will be difficult if not impossible and the digital divide will continue to widen between the rich and the poor. And, all technological solutions come with environmental impact that will inadvertently impact the marginalized communities the most. Thus, when developing technological solutions, we must keep in mind why products are created and for whom, and where technology is designed and deployed, such that we consciously build environmentally-sustainable, socially-responsible, inclusive technologies for the next decades to come. Predicting the innovations of the future is a difficult if not impossible task. However, the political, societal, and environmental challenges the world faces are clear, and technology can play a significant role in helping societies adapt and thrive. The problems and solutions in the three domains of pervasive connectivity, sustainability, and inclusion are interconnected and must be addressed holistically. For instance, adapting the IT infrastructure to climate change must not in turn make climate change worse. Making innovative devices and software to address the needs of an aging population does not mean that poorer populations in other regions can be ignored. And population decline does not fully address sustainability concerns -as standard of living improves with the help of technology, the per-capita environmental footprint of the population also increases [O'Neill et al., 2018] . Technology must be offered that is inclusive of the world's physical, cultural, and economic diversity, and which is manufactured, used, and recycled with environmental sustainability at the forefront. We hope to inspire our architecture community to take broader and more holistic approaches when developing technologies which include a deeper understanding of the biological, societal, cultural, environmental, political, and economic implications of future technological innovations. To summarize: • Build resilient, secure infrastructures -software, hardware, and everything in-between -for the computing spectrum of systems at all sizes and scales. • Design for the entire system life cycle from manufacturing to end-of-life and for the entire software development cycle from experimentation to deployment. Build environmentally-sustainable systems, software, and algorithms -beyond energy efficiency. • Broaden the scope of technologies to serve people from different parts of the world and with differing needs by enabling equitable access to technology and the rich economic opportunities it presents. For the next decades to come, we envision significant cross-disciplinary efforts to build a circular development cycle by placing pervasive connectivity, sustainability, and demographic inclusion at the design forefront in order to sustain and expand the benefits of a technologically rich society. How the pandemic has stress-tested the crowded digital home TMT-Connectivity-and-mobile-trends/DI_TMT-Connectivity-and-mobile-trends.pdf What's Going on With Amazon's "High-Tech" Warehouse Robots? A Growing Presence on the Farm: Robots How the Human Genome Project shook the world of cancer research India's All-Female News Outlet Battles Sexism, Caste -And Hits The Silver Screen Intricacies of medical drones in healthcare delivery: Implications for africa Search and rescue with autonomous flying robots through behavior-based cooperative intelligence Digital technologies in the public-health response to covid-19 Effy Vayena, and Viktor von Wyl. Early evidence of effectiveness of digital contact tracing for sars-cov-2 in switzerland Using AI to help health experts address the COVID-19 pandemic The Story of the Intel 4004 Intel. Intel Core i9-9900KS Processor AMD. AMD Ryzen 9 5950X Desktop Processors Cramming more components onto integrated circuits. Electronics An efficient algorithm for exploiting multiple arithmetic units Decoupled access/execute computer architectures Hpsm, a high performance restricted data flow architecture having minimal functionality A study of branch prediction strategies Branch prediction strategies and branch target buffer design Control flow prediction for dynamic ilp processors Increasing the instruction fetch rate via multiple branch prediction and a branch address cache Dynamic branch prediction with perceptrons Cache memories A case for direct-mapped caches Characteristics of performance-optimal multi-level cache hierarchies Improving direct-mapped cache performance by the addition of a small fully-associative cache and prefetch buffers An effective on-chip preloading scheme to reduce data access penalty Stride directed prefetching in scalar processors Simultaneous multithreading: Maximizing on-chip parallelism Exploiting choice: Instruction fetch and issue on an implementable simultaneous multithreading processor Facebook Sustainability -Data Centers Is PUE actually going UP? Data center PUEs flat since 2013 Hitting the memory wall: Implications of the obvious Design of ion-implanted MOSFET's with very small physical dimensions A 30 year retrospective on Dennard's MOSFET scaling paper Shared memory consistency models: a tutorial Multiprocessors should support simple memory consistency models A Primer on Memory Consistency and Cache Coherence Adaptive insertion policies for high performance caching High Performance Cache Replacement Using Re-Reference Interval Prediction (RRIP) SHiP: Signature-Based Hit Predictor for High Performance Caching Somayeh Sardashti, Angelos Arelakis, Per Stenstom Route packets, not wires: On-chip inteconnection networks Orion: a power-performance simulator for interconnection networks Wattch: A framework for architectural-level power analysis and optimizations Dynamic thermal management for high-performance microprocessors Temperature-aware microarchitecture Computer Architecture Techniques for Power-Efficiency The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines Where the cloud rests: The economic geography of data centers Dark silicon and the end of multicore scaling A domain-specific architecture for deep neural networks Toward accelerating deep learning at scale using specialized hardware in the datacenter Eyeriss: A spatial architecture for energy-efficient dataflow for convolutional neural networks In-datacenter performance analysis of a tensor processing unit Eric Chung, and Doug Burger. A configurable cloud-scale dnn processor for real-time ai Simba: Scaling deep-learning inference with multi-chip-module-based architecture Mlperf: An industry standard benchmark suite for machine learning performance High-Performance Deep-Learning Coprocessor Integrated into x86 SoC with Server-Class CPUs Industrial Product The vision behind mlperf: Understanding ai inference performance Bharath Muthiah, Nitin Nagarkatte, Ashwin Narasimha Sparsity-aware and re-configurable npu architecture for samsung flagship mobile soc Energy efficiency boost in the ai-infused power 10 processor: Industrial product Tensor Cores: Unprecedented Acceleration for HPC and AI Asic clouds: specializing the datacenter for planet-scale applications Warehouse-scale video acceleration: Co-design and deployment in the wild The sharing economy: Why people participate in collaborative consumption Keys under doormats: mandating insecurity by requiring government access to all data and communications On the security and performance of proof of work blockchains Digital privacy challenges with shared mobile phone use in bangladesh Potential for discrimination in online targeted advertising A case study of algorithm-assisted decision making in child maltreatment hotline screening decisions All the cool kids, how do they fit in?: Popularity and demographic biases in recommender evaluation and effectiveness Fairness and abstraction in sociotechnical systems Discrimination through optimization: How facebook's ad delivery can lead to biased outcomes Analyzing biases in perception of truth in news stories and their implications for fact checking Bias in word embeddings Going from bad to worse: from internet voting to blockchain voting On the dangers of stochastic parrots: Can language models be too big? Algorithmic monoculture and social welfare Challenges: Environmental design for pervasive computing systems Energy and policy considerations for deep learning in nlp Examining the Carbon Footprint of Devices Most of computing's carbon emissions are coming from manufacturing and infrastructure Chasing carbon: The elusive environmental footprint of computing The Chip Industry Has a Problem With Its Giant Carbon Footprint 10 Internet Statistics Every Marketer Should Know in 2021 5 steps to get the internet to all americans Two thirds of the world's school-age children have no internet access at home Expanding internet connectivity with stratospheric balloons High-speed, low latency broadband internet Meltdown: Reading kernel memory from user space Spectre attacks: Exploiting speculative execution Azure Storage redundancy Designing for disaster recovery with ExpressRoute private peering Most Texas Data Centers Weathered the Storm, But Things Did Not Go Smoothly Designing And Managing Data Centers For Resilience: Demand Response And Microgrids How an Oregon wildfire almost derailed California's power grid Patterns for resilient architecture -part 1: The story of embracing failure at scale Lighting Up Africa: Bringing Renewable, Off-Grid Energy to Communities Matching renewable energy supply and demand in green datacenters Carbon-aware computing for datacenters. CoRR, abs/2106.11750 Evaluating coupling models for cloud datacenters and power grids Flex: High-availability datacenters with zero reserved power Facebook is committed to reaching net zero emissions across our value chain in 2030, aligning our efforts with the latest science on what is needed to transition to a zero-carbon future Apple commits to be 100 percent carbon neutral for its supply chain and products by 2030 Microsoft will be carbon negative by 2030 The Internet is 24x7-carbon-free energy should be too Amazon and Other Tech Giants Race to Buy Up Renewable Energy China's Pledge to Be Carbon Neutral by 2060: What It Means Climate change: EU to cut CO2 emissions by 55% by 2030 A biodegradable computer chip that performs surprisingly well Architectural tradeoffs for biodegradable computing The Carbon Reduction Opportunity of Moving to Amazon Web Services The Carbon Benefits of Cloud Computing: a Study of the Microsoft Cloud DeepMind AI Reduces Google Data Centre Cooling Bill by 40% Software, servers, systems, sensors, and science: Facebook's recipe for hyperefficient data centers Recalibrating global data center energy-use estimates Durability of smartphones: A technical analysis of reliability and repairability aspects Uptime Institute global data center The world's most sustainable smartphone now at a lower price Why We Must Fight for the Right to Repair Our Electronics What Biden's 'right-to-repair' order could mean for Apple and Tesla Democratizing AI for farmers around the world How to Avoid a Climate Disaster: The Solutions We Have and the Breakthroughs We Need. Knopf, 2021. U.S. Energy Information Adminstration. Electricity explained: Use of electricity An introduction to electrocatalyst design using machine learning for renewable energy storage Machine learning can boost the value of wind energy COVID curbed carbon emissions in 2020 -but not by much Green server design: beyond operational energy to sustainability Dtco including sustainability: Power-performance-area-cost-environmental score (ppace) analysis for logic technologies Lluis-Miquel Munguia, Daniel Rothchild, David So, Maud Texier, and Jeff Dean. Carbon emissions and large neural network training AI for social good: unlocking the opportunity for positive impact Artificial intelligence: Can it help achieve environmental sustainability? Fertility, mortality, migration, and population scenarios for 195 countries and territories from 2017 to 2100: a forecasting analysis for the Global Burden of Disease Study The effectiveness of word prediction software wordq: ". . . predict it, hear it, choose it, review it, correct it, write it now This hidden AirPods feature helps you hear better Voice separation with an unknown number of multiple speakers How VR Is Helping Visually Impaired Patients Regain Close To Normal Levels Of Sight Invited paper: Directly patterened 2645 ppi full color oled microdisplay for head mounted wearables Stanford materials scientists borrow solar panel tech to create new ultrahigh-res OLED display Dynamic temperature management of near-sensor processing for energy-efficient high-fidelity imaging Perspectives on energy storage for flexible electronic systems Current trends in hardware and software for brain-computer interfaces (bcis) Hardware-software co-design for brain-computer interfaces Low-infrastructure methods to improve internet access for mobile users in emerging regions Enabling sustainable sensing in adverse environments Intermittent computing: Challenges and opportunities Beyond the trees: Resilient multipath for last-mile WISP networks. CoRR, abs Dynamic adaptive techniques for learning application delay tolerance for mobile data offloading Energy Department Targets Vastly Cheaper Batteries to Clean Up the Grid Beyond Engagement: Aligning Algorithmic Recommendations With Prosocial Goals The role of cooperation in responsible ai development, 2019. NIST. Nist proposes approach for reducing risk of bias in artificial intelligence Facebook's five pillars of Responsible AI Principles: Recommendations on the Ethical Use of Artificial Intelligence by the Department of Defense Mobile devices are too expensive for billions of people -and it's keeping them offline An Empirical Comparison of Technologically Mediated Advertising in Under-Connected Populations A good life for all within planetary boundaries We would like to thank Doug Carmean, David Brooks, and Lizy John for their insightful feedback on this work.