IEEE Circuits and Systems Magazine - Q3 2022 - 54

Following Prof. Han's talk, Prof. H.-S. Philip Wong
(Willard R. and Inez Kerr Bell Professor, Stanford University)
presented " N3XT-3D-MOSAIC: Domain-Specific
Technology for AI Compute " , noting that 21st century
applications are going to be data-centric. Data analytics,
machine learning, and AI applications are going to
dominate, from data center to mobile and IoT, from collecting
and processing, to curating the data, to deriving
information. Many systems will need to learn and adapt
on the fly. Three-dimensional integration is one of the
major technology directions for integrated circuits. Prof.
Wong gave an overview of the new materials and device
technologies that may need to be developed to realize
monolithic 3D integration with multiple logic transistor
and memory device layers. He gave two examples
of compute-in-memory chips that feature RRAM integration
with CMOS logic as an illustration of how future 3D
systems may be designed.
Dr. Steve Pawlowski (VP at Micron Technology) next
talked about rethinking the memory-compute subsystem.
The best machines struggle on workloads requiring
higher memory and network performance. The
challenge remains to improve overall systems design
starting with memory. Dr. Pawlowski described evolving
usage models driving compute demands that require
vast amounts of data. He suggested current system bottlenecks
will get worse unless something is done. New
memory-compute architectures will have the greatest
impact on systems. However, new architectures will take
time to evolve. Market driven use cases can speed adoption.
Integration of compute into memory to get greatest
energy and performance is key. New devices must build
from and evolve the dominant software ecosystem.
Many future neural networks will require higher bandwidths,
which increases energy consumption. Memory
energy is interconnect dominated. Memory bandwidth
is also pin and locality dominated. A closer coupling of
memory and compute is the path forward.
Subsequently, Dr. Pradeep Dubey (Senior Fellow at
Intel) gave a great talk about the " Era of Ubiquitous AI " .
Artificial intelligence (AI) is touching, if not transforming,
every aspect of our lives. AI is impacting not just
what computing can do for us, rather how computing
gets done. Fast-evolving AI algorithms are driving demand
for general-purpose computing that cannot be met
by " business as usual " engineering. At the same time,
programmers are often data scientists, not computer
scientists; expecting programmers to figure out increasingly
complex hardware on their own just doesn't work.
Architects are therefore needed more than ever-chip
architects to create new processors, systems architects
to design new data centers, software architects to design
new frameworks, and AI architects to churn out
new models and new algorithms. Are we up to the task?
Or do we need to augment human architects with AI to
meet the challenge?
The symposium's concluding talk was given by Tamar
Eilam (IBM Fellow). She presented IBM's initiative related
to sustainable and responsible computing. IBM traditionally
has focused on privacy of data, security, and ethics.
Another key consideration is reducing carbon footprint
(CF). For example, some AI jobs consume CF equivalent
to that of the lifetime of 5 cars. It is predicted that electricity
usage in the data center would increase by 8% by
the year 2030. CF is driving AI to move to cloud. It is crucial
to quantify Carbon Footprint Energy (CFE), which is
defined as product of IT equipment energy, power usage
effectiveness (overhead power conversion and cooling),
and source of the energy (coal, nuclear, etc.). Organizations
need to report CFE to discover hot spots and optimize.
To make greener AI, the use of renewable energy,
controlling the data utilization, and finding alternate
ways to reduce power are essential. In order to do this,
workloads need to be monitored, enabling a breakdown
of CF by cloud tenant and application.
The symposium also featured a poster session, organized
into 3 parallel tracks. Out of 54 posters, the top
3 best posters awarded from each track. The list of winners
is given on the symposium website: https://www.
zurich.ibm.com/thinklab/AIcomputesymposium.html
The symposium closed with a panel discussion on
Responsible Computing, with four distinguished panelists
including Tamar Eilam (IBM), Irena Risch (McGill),
Evgeni Gousev (Qualcomm), and Bouchra Bouqata
(Amazon). Panelists addressed questions spanning a
range of controversial topics, including the environmental
impact of data center workloads, building trust in AI,
eliminating bias in AI, and ensuring data privacy.
Replays of entire two-day symposium are available
on the symposium website: https://www.zurich.ibm.
com/thinklab/AIcomputesymposium.html
Dr. Rajiv Joshi, IEEE Life Fellow
Dr. Arvind Kumar, IEEE Member
Dr. Matt Ziegler, IEEE Member
Affiliation-T. J. Watson Research Center,
Yorktown Heights, NY 10598
Executive Sponsor-Dr. Mukesh Khare
AoT Sponsor-Dr. John " Boz " Handy Bosma
Committee-Rajiv Joshi, Matt Ziegler, Arvind
Kumar, Xin Zhang, Krishnan Kailas, Kaoutar El
Maghraoui, Jin-Ping Han, Anna Topol, John Rozen
54
IEEE CIRCUITS AND SYSTEMS MAGAZINE
THIRD QUARTER 2022
https://www.zurich.ibm.com/thinklab/AIcomputesymposium.html https://www.zurich.ibm.com/thinklab/AIcomputesymposium.html https://www.zurich.ibm.com/thinklab/AIcomputesymposium.html https://www.zurich.ibm.com/thinklab/AIcomputesymposium.html
IEEE Circuits and Systems Magazine - Q3 2022

Table of Contents for the Digital Edition of IEEE Circuits and Systems Magazine - Q3 2022
