ISSCC Videos
ISSCC2020: Plenary - The Deep Learning Revolution and Its Implications for Computer Architecture &..
updated
CSAIL & Andrew (1956) and Erna Viterbi Professor,
Cambridge, MA
In today’s robot revolution, a record 3.1 million robots are now working in factories, doing everything from assembling computers to packing goods and monitoring air quality and performance. A far greater number of smart machines impact our lives in countless other ways—improving the precision of surgeons, cleaning our homes, extending our reach to distant worlds—and we are on the cusp of even more exciting opportunities. Future machines, enabled by recent advances in AI, will come in diverse forms and materials, embodying a new level of physical intelligence. Physical Intelligence is achieved when the power of AI to understand text, images, signals, and other information is used to make physical machines such as robots intelligent. However, a critical challenge remains: balancing the capabilities of AI with sustainable energy usage. To achieve effective physical intelligence, we need energy-efficient AI systems that can run reliably on robots, sensors, and other edge devices. In this paper I will discuss the energy challenges of transformer-based foundational AI models, I will introduce several state space models, and explain how they achieve energy efficiency, and how state-space models enable physical intelligence.
Samsung Electronics, Hwaseong, South Korea
The recent AI revolution, spearheaded by Large Language Models (LLMs), demands substantial computing resources and corresponding memory solutions. However, unlike processors that can leverage advancements in fabrication processes, memory devices are increasingly struggling to meet the high bandwidth, large capacity, and power efficiency requirements of AI systems. This paper analyzes the requirements and limitations of systems in the AI era, categorizing application-specific memory needs in terms of performance, power, and capacity. We introduce performance-
centric solutions such as HBM (High Bandwidth Memory) and PIM (Processing-In-Memory) technologies, energy-efficient solutions including custom HBM and LPW (LPDDR Wide-IO) memory, and capacity-focused solutions like SSD (Solid-State Drives) and CXL (Compute Express Link) Memories. Additionally, we discuss how continuous scaling of DRAM and NAND Flash processes, as well as 3D-packaging technologies, can address the trade-offs among performance, power, and
capacity more effectively. Finally, the importance of software technologies in optimizing the utilization of these increasingly specialized memory solutions is emphasized, along with a discussion of the enabling core technologies for each solution. To meet the high demands of AI systems, the ongoing advancement of existing memory devices and the development of new memory solutions will play crucial roles. These efforts will support the advancement of AI technologies and
contribute to human society.
Infineon Technologies, Munich, Germany
The automotive industry is undergoing a significant transformation, driven by the rise of software-defined vehicles (SDVs). Semiconductors will play a pivotal role in enabling this transition, powering the complex systems that underpin the features and functions of modern cars. This paper explores the key trends driving the growth of the automotive semiconductor market, including green mobility, autonomous driving, and smarter cars. It delves into the challenges and opportunities associated
with the development of SDVs, highlighting the importance of advanced microelectronics, artificial intelligence, and secure communication solutions. The paper concludes by emphasizing the crucial role of semiconductors in shaping the future of mobility. By addressing the challenges and embracing the opportunities presented by SDVs and AI, the automotive industry can create a more sustainable and innovative future.
Foundry Technology Development
Intel, Chandler, AZ
AI holds transformative potential for humanity, enhancing our ability to solve complex problems with speed and accuracy, and unlocking new realms of innovation and understanding. The lightning-fast progression of AI, unprecedented in history, necessitates rapid advancements at a system level, from low-power and edge-AI devices to cloud-based computing, and in the communication networks that connect them. This need for rapid AI system scaling is driving the innovation frontier in silicon, packaging, architecture, and software. This paper describes a matrix of technologies that empowers the industry to achieve remarkable progress at every level, from chips to systems.
Changes with Innovation Around its Constraints
Ian Young, PhD - Intel
This paper provides a comprehensive overview of the future of semiconductor technology, focusing on the interplay between innovation and entrepreneurship. It is organized into sections that discuss the current state of the semiconductor industry, the role of entrepreneurship in driving innovation, and the potential future developments in this field. It highlights the virtuous cycle of innovation and entrepreneurship, where advancements in semiconductor technology fuel new business opportunities, which in turn drive further technological progress. This cycle is crucial for the creation of the next generation of intelligent electronic systems. The paper also explores the ‘stack’ of semiconductor technology, from materials and design to manufacturing and generative AI applications. It establishes that there are beneficial relationships between these different parts of the stack, with advancements in one area often enabling progress in others. This interconnectedness underscores the importance of a holistic approach to innovation in the semiconductor industry, where advancements in different parts of the stack can collectively drive the industry forward.
NVIDIA
Generative AI has captured the imagination of users across multiple industries, and we have only begun to tap the potential of this amazing technology. GenAI applications can create text, computer code, protein sequences, images, video, rendered 3D graphics, music, with more generation types being constantly added. The combination of high compute demands, and the real-time requirements of many GenAI-based applications requires design thinking at datacenter scale. This paper will cover the breadth of technology innovations, from circuits to silicon to software to data center, needed to enable today’s latest supercomputers for GenAI, and discuss our experiences applying GenAI to
various business sectors.
Since its inception, Moore’s Law has been the driving force for IC design. Although during the first decade, “everything” seemed to be better, however, we lost the scaling of processor clock speed and RF transistor speed, and now it looks as if power efficiency of digital gates will stall. What remains is scaling in transistor count and cost-per-function, thanks to 3D integration.
Thus, this is an excellent moment to reconsider how we design for analog and digital signal processing. The higher the required signal-to-noise ratio (SNR), the more power-efficient digital signal processing is compared to analog. Pure analog processing remains more efficient only for ~30dB SNR or less. In the case of digital processing, the conversion from analog to digital should therefore be made as early in the signal chain as possible. Thanks to the figure-of-merit race, analogto-digital converters (ADCs) have experienced a tremendous win in power efficiency. However, these ADCs require a large input voltage swing while the input signals to be converted, from an antenna or sensor interface, are usually much smaller. Therefore, RF and analog front-ends are needed, which consume much more power than the ADCs to be driven.
Let us re-think these analog front-ends. Can we still efficiently design these frontends in future CMOS? Do we need so much linear amplification? Do we need active linear circuits at all? Can we not use “digital” components to replace the analog front-ends and ADCs?
& Overseas Operations Office, Taiwan Semiconductor Manufacturing Company
Semiconductors are the foundation of today’s digital economy and are powering innovations that will shape the trajectory of human history. This paper highlights the latest progress of the semiconductor industry to support a vast spectrum of applications that have forever changed our lives. It gives insight into the paths of continued advanced technology scaling, the essential role of design-technology co-optimization (DTCO), and how system-level integration will elevate system performance to new heights. The advancements of semiconductors will enable many new innovations in artificial intelligence (AI), high-performance computing (HPC), wireless connectivity, and autonomous driving. The paper also provides the trends of technologies ranging from low-power and edge AI devices to cloudbased computing. By harnessing the new capabilities of semiconductors, these innovations will greatly improve productivity, efficiency, safety, as well as
sustainability. The semiconductor industry is indeed experiencing a “golden era” in spurring remarkable economic growth and unleashing innovations to create a better future for society.
In every aspect of our life and society, semiconductors play a major role. The pandemic in conjunction with supply chain hiccups and geopolitical tensions made all regions realize that they need to revisit their presence in the semiconductor value chain. The European Commission projected the ambition of achieving a 20% share of global semiconductor production by 2030.
Europe can leverage existing strengths such as, among others, the unique position of equipment companies and leadership positions in 300mm semiconductor technology R&D. The Chips-for-Europe initiative will invest in pilot lines and ecosystems for chip manufacturing, embracing leading-edge and first-of-a-kind technologies. The pilot lines will allow early exploration of the potential impact of new technology features in advanced chip and system architectures. This will trigger increased demand and accelerate the industrial uptake of novel technologies. This type of innovation loop is also essential for deep-tech start-ups building their unique value proposition. The full-stack, networked model of industry collaboration is at the core of the EU Chips Act ambition and will impact different application domains such as heterogeneous cloud and distributed computing, connectivity, automotive, and health.
It is crucial for all this innovation potential that we, as an industry, consider that semiconductor
manufacturing is resource-intensive with respect to energy, water, chemicals, and raw materials. Design technology co-optimization (DTCO) and System-Technology co-optimization (STCO) methodologies can develop a framework for early sustainability assessments of logic technologies. Finally, we urgently need to get the message across that climate, health, safety, and human connectedness all require complex digital backbones, if we want to stand a chance of attracting the right talent.
Across essentially all industrial sectors, advanced semiconductor technology is the key enabler for
innovations in customer offerings and internal efficiencies. The increase in the value of data and the related push for AI are examples of forces that increase the demand for computing power, which translates to more complex and powerful silicon. Moore’s Law, supported by rapidly evolving semiconductor technology and ever more advanced building practices and assembly technologies, has met the need for decades.
But what is driving 5G today? If we look at the processing requirements, it is the digital front-end, physical layer processing, and beam forming. Back in 2010, LTE/4G was a 20MHz carrier with two receive and two transmit branches, and there was a transmission time interval of one millisecond. Fast forward to where we are today on 5G with massive MIMO, we typically have 100MHz carrier bandwidth. That is a factor of five increase. We have 64 transmitters and 64 receiver radios, which is an increase by a factor of 32, and the transmission time is down to 0.5 milliseconds. In other words, there is only half the time to do 160 times more processing. This is driving an exponential increase in processing needs across the telecom business today and will continue to do so as we race toward 6G. This talk will address whether the semiconductor industry is ready to tackle these challenges.
Although traditional scaling has slowed over the past decade, we have made tremendous progress as an industry with new approaches including chiplet-based architectures, domain-specific accelerators, and advanced packaging technologies which have enabled major milestones including the first exascale supercomputers. As we look into the future, we need to accelerate the pace of innovation to drive the next decade of advancement in high-performance computing. By far, the largest limiting factor to delivering continued compounded growth in computation power is energy efficiency. This paper highlights a holistic strategy for accelerating innovation in energy efficiency required for next-generation high-performance computing and ultimately achieving zetta-scale performance. These approaches will be built on continued innovation in process technologies, modular chiplet architectures, and advanced packaging. Fully meeting the challenge will require new dimensions of improvement through extending domain-specific architectures to accelerate core algorithms in combination with the wide-scale deployment of AI across all aspects of the system from transistors to software.
and CEO of Tech Idea, Kawasaki, Japan
The past 50 years has been an era in which analog equipment has been replaced by digital counterparts. Audio, TV, video, camcorder, camera, recording, wired connection, and wireless communication have been subject to digitization. The digitization of these devices and systems was due to the technological shift from bipolar to CMOS, and to the development of logic and memory circuits supported by scaling laws. In addition, design innovation in mixed-signal integrated circuits such as ADCs and DACs has shown to be indispensable. This talk will look back on the digitization of equipment and the mixed-signal integrated circuit technology that contributed to it. Further, we will look forward to future applications and developments.
0:45:25 - Innovation For the Next Decade of Compute Efficiency
Lisa Su, Chair and Chief Executive Officer, AMD, Austin, TX
1:20:25 - Shape the World with Mixed-Signal Integrated Circuits - Past, Present, and Future
Akira Matsuzawa, Professor Emeritus of Tokyo Institute of Technology and CEO of Tech Idea, Kawasaki, Japan
(not streamed) - ISSCC, SSCS, IEEE Award Presentations
(not streamed) - Break
2:37:25 - EU Chips Act Drives Pan-European Full-Stack Innovation Partnerships
Jo De Boeck, Executive Vice President and Chief Strategy Officer, imec & KU Leuven, Leuven, Belgium
3:15:05 - 5G Drives Exponential Increase in Processing Needs Across all Industries
Erik Ekudden, Senior Vice President & Chief Technology Officer, Ericsson, Kista, Sweden
3:48:03 - Presentation to Plenary Speakers & Conclusion
This event is dedicated to 3rd and 4th-year undergraduate and starting graduate students interested in Circuit Design. Online attendees are encouraged to ask questions during the event (Feb. 18, 7:30am-12:20pm, PST) and we will try to provide answers.
Sensing, artificial intelligence, and actuation will enable autonomous end-to-end system solutions in existing and new application fields including automotive, digital health, agriculture, environmental control, and decarbonization. The Semiconductor Industry is driving this transformation and sensors, smart embedded actuators, analog interfaces, connectivity, security and embedded AI, offer a perfect toolset for companies to continue to innovate. To fuel this innovation, we need to develop energy efficient, high-accuracy, autonomous, ultra-compact, and trusted ICs. These chips need to feature state-of-the-art system and embedded security techniques to protect the gathered data, its processing and the resulting actuation. New and super-efficient computational hardware technologies supporting AI and machine learning are already transforming at-the-edge data processing and are pushing the envelope on intelligent functionality and IoT network scalability.
Future advances will rely on these evolving IC technologies as well as associated packaging solutions. These will include super-integration, wafer-to-wafer bonding, and system-in-package, to enable the heterogenous integration of multiple technologies.
In the follow up to ‘Builders of the Imaginary’, we will unveil the next chapter in autonomous design, piecing together a new breed of super-monolithic devices, dense interconnects, and chiplets, into software-defined, heterogeneous architectures.


