Accelerate Your

AI-Enabled Edge Solution

with Adaptive Computing

Introducing Adaptive
System-on-Modules
(SOMs)
Executive Summary

AI-enabled applications are increasingly being deployed at the edge.

Adaptive SOMs provide a comprehensive, production-ready platform
to build accelerated, AI-enabled applications at the edge, including
smart vision across cities, factories, and hospitals. Such applications
require a large amount of processing to be performed with low
latency, low power consumption and in a small footprint. To
achieve this combination, the whole application (both the AI and
non-AI processes) must be accelerated.

As AI models rapidly evolve, the acceleration platform must

also be adaptable. This allows for optimal implementation
of not just today’s AI techniques, but tomorrow’s as well.
In this eBook, we give an introduction to AI at the edge and
explain how SOMs can be a good production solution.
We also introduce adaptive computing and detail the
benefits that adaptive SOMs bring. Finally we detail
considerations when selecting an AI-enabled
edge solution.
 CHAPTER 1:
THE RISE OF AI-ENABLED EDGE COMPUTING......................................1

CHAPTER 2:
AN INTRODUCTION TO SYSTEM-ON-MODULES (SOMS).......................4

CHAPTER 3:
THE CHALLENGES OF EDGE PROCESSING............................................9

CHAPTER 4:
INTRODUCING ADAPTIVE COMPUTING.............................................. 13

Table of Contents CHAPTER 5:

THE ADAPTIVE SYSTEM-ON-MODULE (SOM).....................................16

CHAPTER 6:
ADAPTIVE SOM BENEFITS FOR HARDWARE DEVELOPERS.................20

CHAPTER 7:
ADAPTIVE SOM BENEFITS FOR SOFTWARE DEVELOPERS..................23

CHAPTER 8:
WHAT TO LOOK FOR WHEN SELECTING AN EDGE SOLUTION.............26

CHAPTER 9:
SUMMARY..........................................................................................30

ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 1:
THE RISE OF AI-ENABLED EDGE COMPUTING

CHAPTER 1

The Rise of AI-Enabled

Edge Computing
SOM USE CASE:
CROP HEALTH ANALYSIS
High-performance AI inference is required
with ultra-low power consumption to maximize
functionality with minimal impact on battery life.

1 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 THE RISE OF AI-ENABLED EDGE COMPUTING

AI-enabled applications are increasingly being is beginning to change with smart retail bringing
deployed at the edge and endpoint. A plethora of automated shopping experiences not thought
applications are being developed that use the latest possible a decade ago. All these applications must
AI techniques across a range of industries and operate with utmost reliability, in harsh conditions,
geographic locations. for 10 years or more. The applications require high
performance, yet must be delivered in an efficient,
Cities are becoming smarter with automated vision reliable, and compact form factor.
applications that manage safety and alert emergency
services when help is needed. Industrial IoT
applications increasingly require high-performance
AI inference processing as factories become smarter
and more automated. Even the retail experience

2 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 THE RISE OF AI-ENABLED EDGE COMPUTING

AI Edge Chipset Revenue, World Markets: 2019-2025

60,000
50,000
Revenue ($m)

40,000
30,000
20,000

10,000
0
2019 2020 2021 2022 2023 2024 2025
Figure 1. AI edge global chipset revenue is forecasted to exceed $50 billion by 2025. (Source: Omdia)

The trend towards increasing AI-enabled edge

computing is highlighted by the market research firm
Omdia, which estimates that global revenue for AI
edge chipsets will surpass $50 billion by 2025.

3 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 2:
AN INTRODUCTION TO SYSTEM ON MODULES (SOM)

CHAPTER 2

An Introduction to
System-on-Modules
(SOMs) SOM USE CASE:
3D PRINTING ARM CONTROL
The 3D printing space is rapidly evolving and
designers are looking to improve both speed and
quality of the process using precise, deterministic
control over a scalable number of axes of motion.

4 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 2: AN INTRODUCTION TO SYSTEM-ON-MODULES (SOMS)

When building an AI-enabled edge application there To save the expense and time of a chip-down
are numerous options available. In many industries, development, design teams may consider utilizing a
it’s common for a hardware design team to perform more integrated solution such as Multi-Chip Module
a “chip-down” development, where specific silicon (MCM), System-in-Package (SIP), Single-Board
devices are chosen, and a fully customized circuit Computer (SBC), or System-on-Module (SOM).
board is developed for the application. While this does
produce a highly optimized implementation, it can
take significant development time and cost to reach
production readiness.

5 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 2: AN INTRODUCTION TO SYSTEM-ON-MODULES (SOMS )

SIP
Stacked Chip 3 Stacked
MCM
Memory Memory
Chip 1 (packaged) Chip 2 (packaged) Chip 1 (unpackaged) Chip 2
Processor
Substrate (single package) Other components Stacked
Substrate (single package) Stacked
Other components
Memory Memory
Circuit board Circuit board

Multi-chip Module (MCM) System-in-Package (SIP)

An MCM combines several (usually packaged) silicon A SIP is like an MCM, except the individual silicon
devices within a single larger package. This has the die are often unpackaged. Some SIPs also have dies
advantage of simplifying circuit board design by that are stacked on top of each other in a 3D array.
reducing the number of individual devices on the SIPs also can include discrete components such as
board. Although there are often advantages in cost resistors and capacitors, reducing the need to include
and performance, MCMs are only available for some these components on the circuit board. SIPs, like
specific applications and still require custom, complex MCMs, still need a fully custom circuit board to be
circuit boards to be developed. designed however – i.e. requiring a full, chip-down
design flow.

6 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 2: AN INTRODUCTION TO SYSTEM-ON-MODULES (SOMS)

SBC SOM

Chip 1 Chip 2 Chip 3

External Chip 1 Chip 2 Chip 3 Power Plug-in board

interfaces supply Other components
Connector Connector
Standalone board
Larger system circuit board

Single-Board Computer (SBC) System-on-Module (SOM)

An SBC is a complete computing system built on a A SOM is a computing system like an SBC but is
circuit board and delivered as a standalone product. designed to connect into a larger solution, rather than
Typically, these include a small microprocessor, be standalone. When built into high-performance
memory and input/output (I/O). SBCs are usually not edge applications, SOMs have many advantages over
built or validated for production deployment. the other solution types mentioned. SOMs provide a
complete, production-ready computing platform, and
save significant development time and cost vs. chip-
down development. SOMs can plug into a larger edge
application, providing both the flexibility of a custom
implementation with the ease-of-use and reduced time-
to-market of an off-the-shelf solution.

7 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 2: AN INTRODUCTION TO SYSTEM-ON-MODULES (SOMS)

“Using a SOM instead of a chip- SOMs are therefore production-ready, proven

down solution will save significant designs that contain all the silicon components
development costs. There is and interface connectors needed for their
shorter development time and specific application area.
fewer mistakes are possible during
development, which both lead to improved time-to-  

market. Furthermore, SOMs come pre-tested to reduce

risk and make it easier for engineers to implement
their custom designs. They don’t have to focus on the
infrastructure that’s on the SOM. Instead, they can
focus on the key differentiators they care about.”

– Bryan Fletcher, Technical Marketing Director at Avnet

8 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 3:
THE CHALLENGES OF EDGE PROCESSING

CHAPTER 3

The Challenges of Edge

Processing
SOM USE CASE:
DIGITAL PATHOLOGY
Typical procedures within radiology, pathology,
dermatology, and ophthalmology are processing
large image sizes requiring complex image
processing, with AI workflows being particularly
compute- and memory-intensive.

9 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 3: THE CHALLENGES OF EDGE PROCESSING

Edge computing is typically limited by power To implement advanced AI applications at the edge,
consumption, footprint, and cost. As processing a domain-specific architecture (DSA) is needed.
demands increase, the challenge of providing the DSAs provide a highly optimized implementation of
performance level required, within the limitations of an application vs. an unaccelerated CPU. They also
edge processing, has increased exponentially. provide determinism and low latency.

CPUs have seen many improvements at the edge, but

the gains have slowed in recent years. Unaccelerated
CPUs struggle to provide the performance needed for
the next generation of AI-enabled edge applications,
especially when considering the tight latency
requirements.

10 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 3: THE CHALLENGES OF EDGE PROCESSING

Whole application acceleration is

WHOLE APPLICATION ACCELERATION

required to implement efficient AI-

enabled applications at the edge.
Performance-Critical
Functions
AI

AI inference needs non-AI pre- and post-processing, Inference

Performance-Critical
all of which have higher performance requirements. A Functions

suitable DSA will be designed specifically to process

the required data efficiently – both the AI inference,
Figure 2. AI Inference is typically part of a larger processing system
and the non-AI parts of the application (i.e. the whole
application). Therefore, whole application acceleration Some ASSPs have been developed for AI edge

is required to implement efficient AI-enabled applications, however, like any fixed silicon solution,

applications at the edge (and elsewhere). they have limitations. Primarily, the pace of AI
innovation is incredibly rapid, rendering AI models
obsolete much quicker than non-AI technologies.
Fixed silicon devices that implement AI can quickly
become obsolete due to the emergence of newer,
more-efficient AI models. It can take several years
to tape-out a fixed silicon device by which time the
state-of-the-art in AI models will have advanced
significantly.

11 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 3: THE CHALLENGES OF EDGE PROCESSING

Security and functional safety requirements are also

becoming more important for edge applications,
often resulting in potentially expensive field updates.
Systems with fixed hardware that cannot adapt to
these changes will quickly become legacy systems.
AI-enabled systems at the edge must be flexible
enough to extend their operating life and maximize
return on investment.

AI Classification
Model Innovation
BN-AlexNet
AlexNet

GoogLeNet

ResNet-18

fflNet 2x w/SE

MobileNet v2
BN-NIN

ENet

VGG-16

ResNet-34

ResNet-50

ResNet-101

ResNet-153

DenseNet-264

SENet-154
VGG-19

Inception-v3

Inception-v4

ResNeXt-101
Fixed Silicon Device 1 Fixed Silicon Device 2

Figure 3. AI models evolve rapidly – much faster than silicon development cycles

12 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 4:
INTRODUCING ADAPTIVE COMPUTING

CHAPTER 4

Introducing Adaptive
Computing
SOM USE CASE:
ROBOTIC CONTROL IN FACTORIES
A true industrial solution supports extended
temperatures, harsh environments, and long
lifecycles in order to be trusted in applications
where downtime is impermissible.

13 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 4: INTRODUCING ADAPTIVE COMPUTING

One of the most-promising technologies for AI- Adaptive computing is more than just hardware,
enabled edge applications is adaptive computing. however. It also encompasses a comprehensive set
Adaptive computing encompasses hardware that can of design and runtime software, that, when combined,
be highly optimized for specific applications such as delivers a unique adaptive platform from which highly
Field Programmable Gate Arrays (FPGAs). In addition flexible, yet efficient systems can be built.
to FPGAs, new types of adaptive hardware have been
recently introduced, including the adaptive System-on-
Chip (SoC) which contains FPGA fabric, coupled with
one or more embedded CPU subsystems.

14 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 4: INTRODUCING ADAPTIVE COMPUTING

Adaptive computing allows the

hardware to be purpose-built for
the application, yet still affords
adaptation as needed, if workloads
or standards evolve.

Adaptive computing allows for DSAs to be

implemented without the design time and upfront cost
needed by custom silicon devices such as ASICs. This
Adaptive computing allows the hardware to be tailored
allows rapid deployment of a flexible, yet optimized
for the application, yet still affords adaptation as
solution for any given domain, including AI-enabled
needed, if workloads or standards evolve.
edge applications.

Adaptive SoCs are ideal for such domain-specific

processing because they combine the flexibility of a
comprehensive, embedded CPU subsystem with the
optimal data processing of adaptive hardware.

15 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 5:
THE ADAPTIVE SYSTEM-ON-MODULE (SOM)

CHAPTER 5

The Adaptive
System-On-Module
(SOM)
SOM USE CASE:
ACCESS CONTROL
AI-powered edge applications like access
control require high-performance, low latency
implementation, but must remain within
limited power and footprint requirements.

16 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 5: THE ADAPTIVE SYSTEM-ON-MODULE (SOM)

As we have seen, SOMs provide a good platform Typical application development teams are
for edge applications. However, to achieve the composed of several types of hardware engineers
performance required by modern AI-enabled and software engineers. Circuit board designers
applications, acceleration is needed. Adaptive are responsible for designing custom boards for
computing provides the acceleration required for AI the required application. In cases where adaptive
applications at the edge. computing (including FPGAs) is used, RTL designers
have traditionally been responsible for configuring
the adaptive devices using Hardware Description
Languages (HDLs) such as Verilog and VHDL. And
software developers write code running on embedded
CPUs using languages such as C++ and frameworks
such as OpenCV, which is commonly used for
embedded vision applications.

17 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 5: THE ADAPTIVE SYSTEM-ON-MODULE (SOM)

Traditionally, all three disciplines are required to build

applications using a custom circuit board, with custom
hardware inside the adaptive SoC, and software
running on the embedded CPU.

As previously mentioned, the design and manufacture

of a custom circuit board containing the adaptive
SoC and other physical components needed by the
application is called ‘chip-down’ development. For
decades, circuit board designers have used chip-
down development to take advantage of adaptive even for vastly different end applications. As
computing. In fact, adaptive SoCs have been deployed industries have moved towards standardized interface
across many AI-enabled edge applications using this and communications protocols, the same set of
method, including in tens of millions of automobiles. components are suitable for a variety of applications,
despite having vastly different processing needs.
Some applications still require custom hardware
components to interface with an adaptive SoC
and thus chip-down design is needed. However, an
increasing number of AI-enabled edge applications
need similar hardware components and interfaces,

18 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 5: THE ADAPTIVE SYSTEM-ON-MODULE (SOM)

An adaptive SOM for an AI-enabled edge application

incorporates an adaptive SoC with industry-standard
interfaces and components, allowing developers with
no or minimal hardware experience to benefit from
adaptive computing technology. An adaptive SoC can
implement both the AI and non-AI processing, thus the
whole application.

An adaptive SoC on an adaptive SOM enables a

high degree of customization, without the need for a
Figure 4. An example of an adaptive SOM
custom PCB. It is designed to be integrated into larger
systems and uses a predefined form factor.

Adaptive SOMs make it possible to take full advantage

of adaptive computing without having to do chip-down
design. An adaptive SOM is just part of the solution,
however. The software is also a critical consideration.
We will cover software considerations in Chapter 7.

19 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 6:
ADAPTIVE SOM BENEFITS FOR HARDWARE DEVELOPERS

CHAPTER 6

Adaptive SOM Benefits for

Hardware Developers
SOM USE CASE:
AUTOMATED GUIDED VEHICLES
Logistics companies require the latest AI models to
enhance productivity and minimize downtime. This must
be delivered in a compact, low power and rugged solution.

20 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 6: ADAPTIVE SOM BENEFITS FOR HARDWARE DEVELOPERS

Typically, hardware developers build AI-enabled edge First, developers need to evaluate and select a device,
hardware by selecting their main silicon devices, build a prototype, and then prove the architecture
then developing an edge form factor board to works sufficiently well with the required software
accommodate the chosen devices. As discussed, this and AI models they plan to deploy. The evaluation
chip-down development can be a costly and complex process alone can take months. Once evaluation
process with a long development cycle. is completed, the production board needs to be
developed, integrated, and manufactured before it
can be deployed.

21 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 6: ADAPTIVE SOM BENEFITS FOR HARDWARE DEVELOPERS

Device-Based RTL/HW Device Board System

Manufacture Deploy
Design Design Debug Design Integration

Module-Based System System Up to 9 Months

Deploy Faster Time to Market
Design Design Integration

Figure 5. SOMs can help accelerate time to market and reduce development cost

Adaptive SOMs allow developers to use an off-the-

shelf, production-ready solution, saving significant
development cost and time. The illustration above
shows how a SOM-based design for an embedded
vision application can be completed up to nine months
faster than a typical chip-down approach.

An adaptive SOM is advantageous for hardware

developers because it allows the design to be changed
late in the design process, vs. a SOM based on fixed-
silicon technology.

22 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 7:
ADAPTIVE SOM BENEFITS FOR SOFTWARE DEVELOPERS

CHAPTER 7

Adaptive SOM
Benefits for Software
Developers
SOM USE CASE:
SURGICAL ROBOTS
Industrial control, communications, machine
vision, machine learning, human-machine
interfaces, cybersecurity, and safety are key
technology considerations.

23 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 7: ADAPTIVE SOM BENEFITS FOR SOFTWARE DEVELOPERS

The advantages of adaptive SOMs are not just limited Recent advancements in software tools, libraries,
to hardware developers. Software developers can and frameworks can enable some design teams to
accelerate their design cycles as well by using pre-built use adaptive computing without burdening hardware
configurations for the underlying adaptive SoCs. engineers. The available comprehensive software
platforms allow software engineers to utilize the
capability of the entire adaptive SoC without needing
specific hardware customization.

24 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 7: ADAPTIVE SOM BENEFITS FOR SOFTWARE DEVELOPERS

Solutions such as accelerated APIs and pre-built

accelerated software platforms can be found freely
available on adaptive SoC vendors’ websites, as
well as from third-party software vendors. Both also
provide access to a range of AI models capable of
performing common AI inference functions.

An adaptive SOM can provide a simple-to-use, out-

of-the-box experience for software developers, from
within familiar environments, including Python, C++,
TensorFlow, and PyTorch. Figure 6. Adaptive SOMs can be programmed from several
abstraction levels

In addition to prebuilt platforms and APIs,

comprehensive software tools enable full
customization of the adaptive hardware, enabling even
more flexibility and optimization.

25 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 8:
WHAT TO LOOK FOR WHEN SELECTING AN EDGE SOLUTION

CHAPTER 8

What to Look for When

Selecting an Edge Solution
SOM USE CASE:
MOTOR CONTROL WIND TURBINES
Wind turbines need to be upgraded remotely
and securely–throughout the world, onshore
and offshore, and over the course of decades.

26 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 8: WHAT TO LOOK FOR WHEN SELECTING AN EDGE SOLUTION

When evaluating a solution for edge applications, allow developers to immediately start application
several factors should be considered: development with a direct path to production
using the same hardware and software. Reduced
Development Time and Cost development costs especially help small and
Successful design teams will look for a turnkey medium-sized companies benefit from adaptive
solution that accelerates development time and computing.
enables fast prototyping. More than just hardware
alone, the solution should consist of an operating
system, a board support package, reference designs,
and open, system-oriented documentation with
easy explanation and guidance. Design cycles can
further be sped up using pre-built starter kits that

27 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 8: WHAT TO LOOK FOR WHEN SELECTING AN EDGE SOLUTION

Performance/Power Ratio applications. Furthermore, the security solution

The solution should provide optimal performance per should be able to adapt over the industrial lifecycle of
watt, which is especially important for applications edge equipment, typically 10-20 years, to keep pace
at the edge typically operating within a pre-set power with evolving threats over that same duration.
envelope of 10W and below. It should deliver domain-
specific architectures with accompanying libraries Future-Proofing
and capabilities for trading off performance and Beyond security, the platform should provide future-
power to suit the needs of a given application. Low proofing to allow both hardware and software
latency performance is not always the same as high- updates, even after the systems are deployed in the
latency performance. For example, GPUs typically field via over-the-air (OTA) updates. The solution
batch data to achieve higher throughput, but at the must be industrial-grade, with a lifecycle of at least
cost of latency. The low-latency throughput is typically 10 years in the field. For similar reasons mentioned
nowhere near as high as peak throughput. with security and the industrial lifecycle, the solution
should accommodate constantly changing AI models
Security as approaches to AI have exploded in number in
The solution should be able to support IEC recent years and will continue to evolve for many
62443-grade cybersecurity, the security standard for years to come. 
the Industrial Internet of Things. This implies, at a
minimum, that the system is built around a hardware-
root-of-trust and includes support for secure and
measured boot with remote attestation of software

28 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 8: WHAT TO LOOK FOR WHEN SELECTING AN EDGE SOLUTION

Ecosystem
“We’ve been recently seeing
The solution should feature a sizable ecosystem,
a lot of interest in SOMs for
offering a variety of services and solutions, including
ML and AI applications. Many
complementary hardware, software, and tools that
customers do not want to get
are pre-integrated with the chosen AI platform or
involved in the specifics of
alternative AI platforms. And, the solution should
complex designs. SOMs allow them to get to
include accelerated application libraries to give
market fast and save development costs.”
software developers the same jumpstart on design
that reference designs do for hardware developers. – Immanuel Rathinam, Associate Director and
Head of R&D and Business Operations;
Adaptive SOMs bring the time-to-market advantage iWave Systems Technologies Pvt Ltd.
of a SOM-based solution, with the application
optimization that can only be achieved with adaptive
computing. This makes it an ideal approach for
implementing AI-enabled edge applications.

29 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 9:
SUMMARY

CHAPTER 9

Summary

SOM USE CASE:

AUTOMATED, MULTI-STORY CAR
PARKING SYSTEM
Reliability is critical in many applications,
with downtime having a high impact on the
business operations.

30 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 9: SUMMARY

Complex, AI-enabled workloads are increasingly SOMs provide an ideal edge processing platform.
moving to the edge. These applications require a When coupled with adaptive SoCs, the resulting
large amount of processing to be performed with adaptive SOMs provide a comprehensive, production-
low latency, low power consumption, and in a small ready platform for AI-enabled edge applications.
footprint. To achieve this, the whole application (both
the AI and non-AI functions) must be accelerated. Companies that move to adaptive SOMs benefit from
a unique combination of performance, flexibility, and
As AI models rapidly evolve, the acceleration rapid development time. They can enjoy the benefits
platform must also be adaptable. This allows optimal of adaptive computing without the need to build their
implementation of not just today’s AI techniques, but own circuit boards, something that has only recently
tomorrow’s as well. been possible at the edge with the introduction of
Xilinx’s Kria™ portfolio of adaptive SOMs.

31 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 CHAPTER 9: SUMMARY

Kria adaptive SOMs are built around Xilinx’s Zynq®

UltraScale+™ MPSoC architecture and give developers
access to a turnkey adaptive computing platform. By
standardizing the core parts of the system, developers
have more time to focus on building in features that
differentiate their technology from the competition.

To learn more about how an adaptive SOM can power

your application, please visit Xilinx’s adaptive SOM
page at: xilinx.com/kria
Figure 7. Kria KV260 Vision AI Starter Kit from Xilinx

Find out more:

Achieving Embedded Design Simplicity with Kria SOMs
Kria K26 SOM: The Ideal Platform for Vision AI at the Edge

About Xilinx:
Xilinx delivers adaptive platforms. Our adaptive SoCs, FPGAs, accelerator cards, and System-on-Modules, give
leading-edge companies the freedom to innovate and deploy, rapidly. We partner with our customers to create
scalable, differentiated and intelligent solutions from the cloud to the edge. In a world where the pace of change is
accelerating, more and more innovators trust Xilinx to help them get to market faster, and with optimal efficiency and
performance. For more information, visit www.xilinx.com.

32 ACCELER ATE YOUR AI-ENABLED EDGE SOLUTION WITH ADAPTIVE COMPUTING

 Corporate Headquarters
Xilinx, Inc.
2100 Logic Drive
San Jose, CA 95124
USA
Tel: 408-559-7778
www.xilinx.com

Xilinx Europe
Bianconi Avenue
Citywest Business Campus
Saggart, County Dublin
Ireland
Tel: +353-1-464-0311
www.xilinx.com

Learn more about:

Japan
Xilinx K.K.
Art Village Osaki Central Tower 4F

Xilinx Kria adaptive SOMs 1-2-2 Osaki, Shinagawa-ku

Tokyo 141-0032 Japan
Tel: +81-3-6744-7777
japan.xilinx.com

Get in touch with us! Asia Pacific Pte. Ltd.

Xilinx, Asia Pacific

[email protected] 5 Changi Business Park

Singapore 486040
Tel: +65-6407-3000
www.xilinx.com

India
Xilinx India Technology Services Pvt. Ltd.
Block A, B, C, 8th & 13th floors,
Meenakshi Tech Park, Survey No. 39
Gachibowli(V), Seri Lingampally (M),
Hyderabad -500 084
Tel: +91-40-6721-4747
www.xilinx.com

© Copyright 2021 Xilinx, Inc. Xilinx, the Xilinx logo, Artix, ISE, Kintex, Kria, Spartan, Versal, Virtex, Vitis, Vivado, Zynq, and other designated
brands included herein are trademarks of Xilinx in the United States and other countries. AMBA, AMBA Designer, ARM, ARM1176JZ-S,
CoreSight, Cortex, and PrimeCell are trademarks of ARM in the EU and other countries. PCIe, and PCI Express are trademarks of PCI-SIG
and used under license. All other trademarks are the property of their respective owners.
Printed in the U.S.A. WW4/9/21