IEEE Solid-States Circuits Magazine - Fall 2021 - 39

used to produce graphene sheets,
such as CVD [39]. High-voltage devices
based on gallium nitride (GaN) and
silicon carbide are gaining interest.
GaN high-electron mobility transistors
lower specific on-resistance and
output capacitance as well as the lack
of reverse-recovery losses enable efficient
voltage conversion over a wide
input voltage range up to 650 V and
beyond [40]. Silicon carbide MOSFETs
bring the advantages of the innovative
WBG materials to your next design as
they offer an extended range of voltage,
rating from 650 to 1,700 V [41].
FinFET transistor technology has enabled
the development of IC technology
to continue to follow Moore's law
[42]. Gate-all-around FET GAAFET is an
evolution of the FinFET structure that
is used in most of today's high-end
process technologies, providing users
with a four-gate structure around the
transistor's channels. A multibridgechannel
FET is similar to a GAAFET, except
for the use of nanosheets instead
of nanowires [43].
New Circuit- and SystemImplementation
Techniques
Machine learning can help speed up
the slow and extensive analog circuit
design process and provide an even
more optimal design. The problem,
however, with conventional supervised
learning techniques is that a
sufficiently large data set is not readily
available due to the slow speed
of simulations in addition to each
design being custom designed for
a specific intellectual property. Reinforcement
learning serves as an
alternative to supervised learning
where the algorithm learns how to
design optimum circuits [44]. Additionally,
a graph convolutional neural
network can be used to transfer the
learned knowledge for transistor sizing
to different topologies as well as
technology nodes paving the way for
design automation and smooth integration
between system blocks [45].
Recent works have also proposed the
use of machine learning to automate
and speed up the design process of
microwave circuits; for example, the
work in [46] proposes a graph neural
network that learns how to simulate
electromagnetic structures.
The separation of data storage and
processing units in a von Neumann architecture
has become a bottleneck for
deep learning applications with high
data-access rates [47]. In-memory computing,
which brings processing inside
the memory and enables multiply
and accumulate operations to be performed
directly on the stored weights,
can reduce computation energy significantly.
The candidates for in-memory
computation include both charge- and
resistance-based memory technologies,
including static random-access
memory (RAM) [48], dynamic RAM [49],
and flash [50] in the former category,
along with phase-change memory [51]
and resistive RAM [52] in the latter.
Emerging workloads such as cloud
computing and big data analytics have
pushed product design to embrace a
more holistic approach for optimization
at the system level. New packaging
technologies offer the freedom and
advantage to design a product more
holistically as a system of minichips
that offers key advantages versus designing
a larger monolithic die [53].
φ1
Vin
Cfly
φ1
φ2
(a)
3
Simulation
2.5
2
1.5
1
0.5
Measurements
0 246 810
ILoad (µA)
(c)
FIGURE 17: (a) The schematic, (b) micrograph, and (c) measurement results of the fabricated
switched-capacitor dc-dc converter [32]. GND: ground.
IEEE SOLID-STATE CIRCUITS MAGAZINE
FALL 2021
39
fsw = 10 kHz
fsw = 50 kHz
fsw = 100 kHz
Cout
Vin
φ2
Vout
φ1
FET FET
GND
Cfly
(b)
New Silicon-Based Applications
Optogenetics is an application that
can benefit from the combination of
electronics and photonics. It is a technology
that allows the targeted, fast
control of precisely defined events
in biological systems as complex as
freely moving mammals [54]. Holography
is another application for
combining electronics and photonics.
It is a method of recording and
reproducing 3D images; however, the
technology has not been widely used
in practical applications because a
hologram requires an enormous volume
of data [55].
Automotive electronics is also
another exciting application. The electronic
components are increasing in
automotive applications, replacing earlier
mechanical and hydraulic solutions
[56]. They also incorporate advanced and
robust diagnostic circuits that provide
information a priori for safety to drivers
and passengers in cars about load
conditions, faults, and so on; such components
are referred to as smart power
circuits [57]. Quantum computing is expected
to extend the high-performance
computing road map if a large number
of quantum bits (qubits) are operable.
φ2
Cout
Vout
Vout (V)

IEEE Solid-States Circuits Magazine - Fall 2021

Table of Contents for the Digital Edition of IEEE Solid-States Circuits Magazine - Fall 2021