8.1 Special Day Session on Future and Emerging Technologies: NanoSystems: Connecting Devices, Architectures, and Applications

Time	Label	Presentation Title Authors
17:00	8.1.1	3D NANOSYSTEMS: THE PATH TO 1,000X ENERGY EFFICIENCY Author: Max Shulaker, MIT, US Abstract While trillions of sensors connected to the "Internet of Everything" (IoE) promise to transform our lives, they simultaneously pose major obstacles which we are already encountering today. The massive amount of generated raw data (i.e., the "data deluge") is quickly exceeding computing capabilities of existing systems, and cannot be overcome by isolated improvements in sensors, transistors, memories or architectures alone. Rather, an end-to-end approach is needed, whereby the unique benefits of new emerging nanotechnologies - for sensors, memories and transistors - are exploited to realize new nanosystem architectures that are not possible using today's technologies. However, emerging nanomaterials and nanodevices suffer from significant imperfections and variations. Thus, realizing working circuits, let alone transformative nanosystems, has been infeasible. In this talk, I present a path towards realizing future nanosystems, and show how recent progress in several emerging nanotechnologies (carbon nanotubes for logic, non-volatile memories for data storage, and new materials for sensing) enables us to realize such nanosystems today. As a case-study, I will discuss how by leveraging emerging nanotechnologies, we have realized the first monolithically-integrated three-dimensional (3D) nanosystem architectures with vertically-integrated layers of logic, memory, and sensing circuits. With dense and fine-grained connectivity between millions of on-chip sensors, data storage, and embedded computation, such nanosystems can capture terabytes of data from the outside world every second, and produce "processed information" by performing in-situ classification of the sensor data using on-chip accelerators. As a demonstration, we tailor a demo system for gas classification, for real-time health monitoring from breath.
17:30	8.1.2	RESISTIVE RAM FOR NEW COMPUTING SYSTEMS: FROM DEEP LEARNING TO BIOMIMICRY Author: Elisa Vianello, CEA LETI, FR Abstract Resistive random-access memory (RRAM) is a memory technology that promises high-capacity, non-volatile data storage, low voltages, fast programming and reading time (few 10's of ns, even <1ns), single bit alterability, execution in place, good cycling performance (higher than Flash), density. Moreover RRAM can be easily integrated in the Back-End-Of-Line of advanced CMOS logic. This will revolutionize traditional memory hierarchy and facilitate the implementation of in-memory computing architectures and Deep Learning accelerators. To further improve the connectivity between memory arrays and computing, a combination of logic 3D Sequential Integration (3DSI) and memory arrays is a promising solution. Thanks to low processing thermal budget (<400°C), thermal stability (>500°C) and low cost (few additional masks), RRAM technologies are good candidates to be inserted in between sequentially stacked MOSFET tiers. RRAMs are also promising candidates for implementing energy-efficient bioinspired synapses, creating a path towards online real time unsupervised learning and life-long learning abilities. We will also explore the use of RRAM for future circuits and systems inspired by the emerging paradigm of biomimicry.
18:00	8.1.3	HOW MIGHT NEW TECHNOLOGIES FOR SENSING SHAPE THE FUTURE OF COMPUTING Author: Naveen Verma, Princeton University, US Abstract The tremendous value computation has shown across applications is driving its expansion from cyber systems to systems that pervade every aspect of our lives. This is being fueled especially by algorithms from artificial intelligence, leading to systems qualified for such integration in our lives, with cognitive capabilities approaching those of humans. A fascinating consequence for system designers is that a tight coupling now results between the data sensed from the physical world and the computations performed on that data. This enforces a unification of design spaces, where new sensing technologies open up new algorithmic opportunities, which in turn open up new architectural options, bringing the potential to overcome traditional bottlenecks in computing. But, a conceptual unification is not enough, a technological unification is also needed. This talk explores such a unification, via hybrid systems based on Large-Area Electronics (LAE) and silicon-CMOS technologies. LAE enables diverse, expansive, and form-fitting sensors, which can be associated with physical objects. This yields sematic structure in the sensor data, which can be exploited towards simpler machine-learning models that are both more data efficient, in terms of learning, and specifically well suited for energy-aggressive hardware architectures. Illustrations will be presented based on integrated LAE-CMOS sensory-compute system prototypes.
18:30		End of session

Time

Label

Presentation Title
Authors

17:00

8.1.1

3D NANOSYSTEMS: THE PATH TO 1,000X ENERGY EFFICIENCY
Author:
Max Shulaker, MIT, US
Abstract
While trillions of sensors connected to the "Internet of Everything" (IoE) promise to transform our lives, they simultaneously pose major obstacles which we are already encountering today. The massive amount of generated raw data (i.e., the "data deluge") is quickly exceeding computing capabilities of existing systems, and cannot be overcome by isolated improvements in sensors, transistors, memories or architectures alone. Rather, an end-to-end approach is needed, whereby the unique benefits of new emerging nanotechnologies - for sensors, memories and transistors - are exploited to realize new nanosystem architectures that are not possible using today's technologies. However, emerging nanomaterials and nanodevices suffer from significant imperfections and variations. Thus, realizing working circuits, let alone transformative nanosystems, has been infeasible. In this talk, I present a path towards realizing future nanosystems, and show how recent progress in several emerging nanotechnologies (carbon nanotubes for logic, non-volatile memories for data storage, and new materials for sensing) enables us to realize such nanosystems today. As a case-study, I will discuss how by leveraging emerging nanotechnologies, we have realized the first monolithically-integrated three-dimensional (3D) nanosystem architectures with vertically-integrated layers of logic, memory, and sensing circuits. With dense and fine-grained connectivity between millions of on-chip sensors, data storage, and embedded computation, such nanosystems can capture terabytes of data from the outside world every second, and produce "processed information" by performing in-situ classification of the sensor data using on-chip accelerators. As a demonstration, we tailor a demo system for gas classification, for real-time health monitoring from breath.

17:30

8.1.2

RESISTIVE RAM FOR NEW COMPUTING SYSTEMS: FROM DEEP LEARNING TO BIOMIMICRY
Author:
Elisa Vianello, CEA LETI, FR
Abstract
Resistive random-access memory (RRAM) is a memory technology that promises high-capacity, non-volatile data storage, low voltages, fast programming and reading time (few 10's of ns, even <1ns), single bit alterability, execution in place, good cycling performance (higher than Flash), density. Moreover RRAM can be easily integrated in the Back-End-Of-Line of advanced CMOS logic. This will revolutionize traditional memory hierarchy and facilitate the implementation of in-memory computing architectures and Deep Learning accelerators. To further improve the connectivity between memory arrays and computing, a combination of logic 3D Sequential Integration (3DSI) and memory arrays is a promising solution. Thanks to low processing thermal budget (<400°C), thermal stability (>500°C) and low cost (few additional masks), RRAM technologies are good candidates to be inserted in between sequentially stacked MOSFET tiers. RRAMs are also promising candidates for implementing energy-efficient bioinspired synapses, creating a path towards online real time unsupervised learning and life-long learning abilities. We will also explore the use of RRAM for future circuits and systems inspired by the emerging paradigm of biomimicry.

18:00

8.1.3

HOW MIGHT NEW TECHNOLOGIES FOR SENSING SHAPE THE FUTURE OF COMPUTING
Author:
Naveen Verma, Princeton University, US
Abstract
The tremendous value computation has shown across applications is driving its expansion from cyber systems to systems that pervade every aspect of our lives. This is being fueled especially by algorithms from artificial intelligence, leading to systems qualified for such integration in our lives, with cognitive capabilities approaching those of humans. A fascinating consequence for system designers is that a tight coupling now results between the data sensed from the physical world and the computations performed on that data. This enforces a unification of design spaces, where new sensing technologies open up new algorithmic opportunities, which in turn open up new architectural options, bringing the potential to overcome traditional bottlenecks in computing. But, a conceptual unification is not enough, a technological unification is also needed. This talk explores such a unification, via hybrid systems based on Large-Area Electronics (LAE) and silicon-CMOS technologies. LAE enables diverse, expansive, and form-fitting sensors, which can be associated with physical objects. This yields sematic structure in the sensor data, which can be exploited towards simpler machine-learning models that are both more data efficient, in terms of learning, and specifically well suited for energy-aggressive hardware architectures. Illustrations will be presented based on integrated LAE-CMOS sensory-compute system prototypes.

18:30

End of session