NEWS EDGE MACHINE LEARNING
Edge learning breakthrough
CEA-LETI REPORTS MACHINE LEARNING
BREAKTHROUGH EXPLOITING
RESISTIVE-RAM TO CREATE INTELLIGENT
SYSTEMS. NEIL TYLER REPORTS
CEA-Leti scientists have demonstrated a machinelearning
technique that exploits the “non-ideal”
traits of resistive-RAM (RRAM) devices, and in the
process overcoming barriers to developing RRAMbased
edge-learning systems.
In a paper published in the January issue
of Nature Electronics, the research team
demonstrated how RRAM, or memristor,
technology can be used to create intelligent
systems that learn locally at the edge,
independent of the cloud.
The learning algorithms used in current RRAMbased
edge approaches cannot be reconciled
with device programming randomness, or
variability, as well as other intrinsic non-idealities
of the technology.
To get around this, the team developed a
method that actively exploits that memristor
randomness, implementing a Markov Chain
Monte Carlo (MCMC) sampling learning algorithm
in a fabricated chip that acts as a Bayesian
machine-learning model.
While machine learning provides the enabling
models and algorithms for edge-learning
systems, increased attention concerning how
these algorithms map onto hardware is required
in order to be able to bring machine learning to
the edge. Machine-learning models are normally
trained using general purpose hardware based on
a von Neumann architecture, which is unsuited
for edge learning because of the energy required
to continuously move information between
separated processing and memory centres onchip.
The microelectronics industry is currently
focused on using RRAM as non-volatile analogue
devices in hardware-based arti cial neural
networks that can allow computation to be
carried out in-memory, drastically reducing energy
requirements. RRAM has been applied to inmemory
implementations of backpropagation
algorithms in order to implement in-situ
learning on edge systems. However, because
backpropogation requires high-precision memory
elements, previous work has largely focused on
how RRAM randomness can be mitigated.
The CEA-Leti team used an approach that
leveraged this randomness, allowing in-situ
learning to be realised in a highly ef cient fashion
through the application of nanosecond voltage
pulses to nanoscale memory devices resulting
in a very low-energy solution. According to the
team, relative to a CMOS implementation of its
algorithm, this approach requires ve orders of
magnitude less energy.
Wi-Fi 6E
Tri-Band chipset
Distribution partnership for Coral platform
OKdo is now supplying a full portfolio of products
from Coral, Google’s platform for edge AI.
Coral offers a comprehensive platform for the
development of devices that can leverage local
arti cial intelligence (AI) using power-ef cient
edged-based tensor processing units (edge
TPUs). This means that constant energy-sapping
Internet connectivity is not required, plus there
are no latency or network security issues.
Comprising hardware components, software
tools and related content, the Coral product
portfolio is fully compatible with Raspberry Pi
and allows emerging application opportunities
to be addressed in healthcare, environment
monitoring, smart agriculture, industrial
automation and vision systems.
Among the platform’s highlights are: the
Coral Dev Board; Coral System-on-Module (SoM);
Coral USB Accelerator; Coral Camera; Coral
Environmental Sensor Board and the Coral Mini
PCIe Accelerator.
“Coral is becoming increasingly prevalent
across industries as more companies discover the
bene ts of edge AI, which brings stronger privacy
and better performance at low cost and low
power to a diverse set of use cases,” explained
Billy Rutledge, Director of Coral at Google. “By
engaging with OKdo, we are well positioned to
serve prospective customers all over the world,
from hobbyists to big corporations.”
According to NXP Semiconductors, its
CW641 Wi-Fi 6E Tri-Band system-on-chip
(SoC) will help to lay the foundation for
a new era of Wi-Fi 6 devices that can
operate in the 6GHz band.
The Wi-Fi 6E device makes use of the
6GHz band and extends Wi-Fi capacity
by bringing higher throughout, increased
capacity, reliability, and improved
latency.
Designed for access points and
service provider gateways, the CW641 is
intended to unlock increased speeds of
over 4Gbps and multi-user performance
in the 6 GHz band, providing greater
capacity and lower latency, and
improving the Wi-Fi user experience.
Adding 6GHz capabilities to gateway
platforms will give service providers
options to effi ciently partition available
bandwidth across devices to ensure
optimum user experience for a wide
range of applications.
Mission critical, high bandwidth, low
latency applications like mesh back haul
and cloud gaming are seen as being
suitable for migration to 6GHz, freeing
up the 5GHz and 2.4 GHz bands for other
lower bandwidth applications.
Beyond access point applications,
the CW641 SoC will enable high
performance Wi-Fi 6E applications across
consumer, automotive, industrial, and
Internet of Things (IoT).
“The Wi-Fi 6E chipset combines
multi-gigabyte data rates, low latency,
and higher multi-user performance to
deliver on customer demands for 6GHz
products that address the decadelong
need for the greater capacity
required in today’s wireless networking
applications,” said Larry Olivas, Head
of Marketing for NXP’s Wireless
Connectivity Solutions.
www.newelectronics.co.uk 26 January 2021 7
/www.newelectronics.co.uk