Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
FROST & SULLIVAN BEST PRACTICES AWARD
New Product Innovation Award 2019AI Accelerator Chips - NORTH AMERICA
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 2 “We Accelerate Growth”
Contents
Background and Company Performance ........................................................................ 3
Industry Challenges .............................................................................................. 3
New Product Attributes and Customer Impact .......................................................... 3
Conclusion........................................................................................................... 4
Significance of New Product Innovation ......................................................................... 8
Understanding New Product Innovation ......................................................................... 8
Key Benchmarking Criteria .................................................................................... 9
Best Practices Award Analysis for Gyrfalcon Technology Inc ............................................. 9
Decision Support Scorecard ................................................................................... 9
New Product Attributes ....................................................................................... 10
Customer Impact ............................................................................................... 10
Decision Support Matrix ...................................................................................... 11
Best Practices Recognition: 10 Steps to Researching, Identifying, and Recognizing Best Practices ................................................................................................................. 12
The Intersection between 360-Degree Research and Best Practices Awards ..................... 13
Research Methodology ........................................................................................ 13
About Frost & Sullivan .............................................................................................. 13
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 3 “We Accelerate Growth”
Background and Company Performance
Industry Challenges
The IoT (internet of things) industry has undergone radical changes in recent years, with
the constant evolution of groundbreaking technologies. In particular, artificial intelligence
(AI) and Big Data have been key enablers for the rapid development of the connected and
smart devices industry as they bring in a new dimension to devices by making them
intelligent. Frost & Sullivan research shows the technology industry is keen on integrating
AI into electronic devices to leverage benefits, such as sharpening machine hearing,
speech, and vision; accessing the growing volumes of stored data; and enabling devices to
learn from experience.
The research on AI is a global phenomenon that is being ardently pursued by global
universities and research institutes, such as Stanford and Cambridge. AI’s initial
developments centered around academic research include a few theories on AI
implementation cases that were demonstrated using mainframe systems and graphic
processing units (GPUs); however, this AI research was not feasible to be carried out on a
larger scale or transpired into a business venture. The introduction of convolutional neural
networks (CNN) revolutionized the AI industry; however, the AI ecosystem had yet to
evolve. Chipsets consumed more power and time for processing AI algorithms and
occupied significant space in the circuits, making integration into electronic devices, such
as smartphones, difficult.
Another important criterion to consider is the cost involved in commercializing chips. To
achieve high adoption, developed chips must cater to millions of users or IOT endpoints,
where they are deployed in bulk quantities. Furthermore, the AI industry is coming from
the orientation of GPUs and mainframes and relies heavily on AI experts or PhDs to create
algorithms, thereby stalling AI’s evolution because advanced AI hardware and software
platforms typically require specialized training for professionals to optimize their use in
specific applications.
Founded in 2017 by a team of AI, memory, and semiconductor veterans, California-based
Gyrfalcon Technology Inc (GTI) has created a domain-specific architecture that can be
optimized for AI applications and integrated into mature process technologies to reduce
the cost of producing AI chips, while attaining industry-level performance with reduced
energy consumption. In 2016, the company’s founders’ initial proof of concept used an
off-the-shelf field-programmable gate array (FPGA). GTI has a global presence, with
offices in China, Japan, and Korea for catering to the AI needs of customers worldwide.
Within a year of inception, GTI successfully released the first of its Lightspeeur© AI
Accelerator chip, followed by two more in 2018. These chips can be integrated into a
variety of smart devices and cloud platforms, thereby paving the way for the next
generation of AI devices.
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 4 “We Accelerate Growth”
New Product Attributes and Customer Impact
Design and Quality
After the initial success achieved by demonstrating the proof of concept on a FPGA
system, GTI focused on designing chipsets for optimizing AI. With the vision to make AI
efficiently deliver high performance with low energy consumption, the company invested
in creating proprietary technologies to enable the development AI accelerator chips that
can be integrated into smart devices. To address the design constraints evident in the AI
industry, GTI incorporated the unique approach of using their proprietary Matrix
Processing Engine (MPE) for processing two-dimensional (2D) data, which mainly
comprises images or voice data, and utilizing its proprietary Constreaming™ Engine
technology. As a result, the input 2D data is maintained in its original form because it is
rapidly processed through the matrix architecture without wasting any energy to convert
it, unlike time- and energy-consuming GPUs and other AI chip methods that use vector
technology for data conversion.
Another element of the MPE is GTI’s proprietary AI
processing in memory (APiM) that integrates memory
and logic into a single chip. This technology improves
performance and reduces energy consumption by
avoiding any unnecessary data movement between the
processor and discrete memory units. This
breakthrough chip 2801 is the first accelerator chip in
GTI’s Lightspeeur portfolio and performs with 9.3 tera
operations per seconds per watt (TOPs/watt) with a
peak performance of 5.6 TOPs at 100 MHz, which can
also consume power as low as 300 milliwatts (mW).
The size of this ASIC chip is 7 mm x 7 mm using 28 nm process technology.
The second chip launched was the 2803, which was optimized for advanced edge and data
center use cases, such as Cloud AI operations. It delivers 24 TOPs/watt and consumes a
minimum power of 700 mW. It can be offered in a multiple chip configuration (16 per PCIe
card) to deliver 270 TOPS while using only 28 watts of power. Both chips are cost
effective, energy efficient, compact, and compare favorably against other AI chip solutions
which require a more circuit area and consume much more power, while delivering lower
performance in terms of TOPs. Therefore, GTI’s chips are ideal for running AI applications,
such as audio and video processing.
The third chip launched, the 2802 is yet another innovation from GTI, in collaboration with
its semiconductor fabrication partner Taiwan Semiconductor Manufacturing Company
Limited (TSMC), released in December 2018. GTI developed its proprietary Gyrfalcon
MRAM engine (GME) to be the first AI chip in the industry with integrated memory
magnetic random access memory (MRAM). As more AI accelerator solutions are trying to
pair memory and logic, this is a key advantage to GTI’s set of technologies and designs.
Lightspeeur© AI
Accelerator Chip 2801 Image Source: GTI
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 5 “We Accelerate Growth”
With the built-in non-volatile memory of 40 megabytes (MB), this chip can run large AI
models, or the memory can be partitioned to run concurrent models on a single chip.
Using non-volatile memory enables the deployment of applications in remote areas and
can be powered by alternative energy generation technologies, such as solar. In addition,
the memory does not need any lead time to load the models or any other information
pertaining to the AI application, which is a great advantage for IoT endpoints.
GTI has a strong portfolio of IP, with more than 50 patents, awarded or in process,
protecting its technology against infringement.
GTI’s AI accelerator chips can be integrated into host processes, thereby boosting the
performance of AI applications. For example, AI data coming from various device sensors
invariably goes to the host processor that runs the application. The data is then delivered
to the AI accelerator for further processing based on the model embedded on the chip,
and the calculated result is returned to the application running on the host device. Based
on application requirements, models can be designed and embedded in the AI accelerator
chip and can be replaced based on the application running on the host device.
GTI uses the 22 nm processing technology, which can be implemented in many
semiconductor fabrication facilities to create chipset
modules with high performance and low energy chips
deployed for an edge-AI product.
Frost & Sullivan commends GTI’s ability to develop a
cutting-edge technology for creating AI accelerator chips
that can be integrated into a variety of host processors
embedded in electronic devices, such as smartphones,
tablets, and industrial robots.
Reliability
Software algorithms are improving every year, and GTI addresses this rapid change by
providing its AI chips with a flexible architecture so that it can be deployed to the neural
net topology of Caffe, TensorFlow, and PyTorch, which are the three most widely used AI
development platforms. Another advantage of accelerator chips is their ease of integration
into the host processor of any company; therefore, these chips are ideally suited for cloud
applications, in addition to edge-AI applications. Moreover, by integrating multiple 2803
chips (16 typically) into the peripheral component interconnect express (PCIe) card, along
with the host processor, multiple AI applications can be run in parallel.
Compared to other AI accelerator chips in the market, GTI’s chips offer ten times more
efficiency in terms of performance to power usage, which can drastically reduce
datacenter costs. The PCIe card integrated with the 2803 chip delivers a performance of
270 TOPs and consumes only 28 watts, which is high compared to other AI-based PCIe
card vendors that offer either 65 TOPs at 70 watts or 248 TOPs at 300 watts. In addition,
the thermal footprint of GTI’s chips are more effective than other chips because they do
not generate as much heat, thereby eliminating any need for additional cooling
equipment. Cascading is another unique feature to the 2803 chip, allowing large AI
8 Chip PCIe Card
Image Source: GTI
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 6 “We Accelerate Growth”
models to be run in datacenters across a limitless number of 2803 chips that have been
embedded in the datacenter rack.
Based on Frost & Sullivan analysis, GTI’s AI accelerator chips exhibit high reliability in
extreme data-centric operations because performance is not compromised, and energy
consumption is kept at a minimum. Furthermore, the cascading architecture enables the
modeling of complex AI applications, which is not feasible with existing competing chips.
Customer Ownership Experience and Customer Service Experience
The launch of GTI’s 2801 chip created a new wave in the AI industry, which had been
bogged down by complex GPUs and mainframe systems. The 2801 chip garnered
immediate interest from Tier I companies, such as Samsung and LG, which have been
actively creating next-generation devices, with a heavy focus on AI. GTI’s AI accelerator
chips are analogous to the bricks of a building because they form key building blocks for
constructing any kind of AI application. The core functionality of these AI accelerator chips
is the processing of input AI data, irrespective of its format (e.g., visual or audio), using
GTI’s patented technologies MPE and APiM and running the data through the chip
embedded with a model optimized for a particular application. The chips can boost any
application, such as AI thermostats, sensor hubs, drones, and robots. Regardless of
application, the core technology executes in the same way, thus providing customers with
a sense of ease in deploying the chips, without undergoing any design changes. In
addition, GTI supports its clients in modifying the accelerator’s operating parameters, such
as higher TOPS for one chip running bigger models.
One of the key challenges in the AI industry is the lack of awareness in the developer
community because of the complexity involved in creating an AI ecosystem. With the goal
of making its products more accessible, GTI offers development kits to customers to help
them create simple AI models and provide a first-hand experience. The proprietary
universal serial bus (USB) stick called Plai™ Plug embedded with a 2801 chip, along with
the Plai™Builder software, provides customers with the ability to create independent
model designs and facilitate the development of a proof of concept. The ability to provide
requisite tools for creating independent model designs is in line with GTI’s goal of enabling
customers to be independent and masters of their own models, while continuing to
provide the building blocks in the form of AI chips. Moreover, GTI supports
customers in designing custom chips if they want dedicated chips with specific
parameters pertaining to their applications.
Apart from Samsung and LG, GTI’s clientele consists of several prominent industry
leaders, such as Foxconn, Alibaba Group, Socionext, ArcSoft, and Konica Minolta. In
collaboration with GTI, Socionext, a global system-on-chip (SoC) developer, has created
an AI server solution by integrating its proprietary parallel multi-core processor
SynQuacer™ SC2A11 with GTI’s AI accelerator chips, which are leveraged for developing a
video and image analysis system for surveillance, live streaming, and other video
applications. QuickLogic is considering the possibility of leveraging GTI’s AI chips in its
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 7 “We Accelerate Growth”
next-generation sensor hubs used for recognizing wake words in voice recognition-capable
gadgets. LG is expected to release its AI products embedded with GTI’s chips in 2019.
GTI’s value proposition is to optimize the integration of AI into chip designs to deliver the
best ratio of high performance-to-low energy consumption with reduced production costs.
Compared to competing solutions, GTI’s technology offers more design flexibility because
the customer can mix and match whatever chips they want based on the end application.
For example, when building a robot, one 2801 chip can be used to make a simple robot, or
several 2801 chips can be packaged into a card and used to enhance capabilities by
running multiple functions in parallel. Moreover, GTI’s accelerator chips aggregate the
processing power and allow customers to conduct advanced edge and cloud data
computing operations/implementations.
Frost & Sullivan recognizes GTI’s successful ability to address the emerging demand for AI
applications through its series of AI accelerator chips and its ability to enable customers to
develop AI applications cost effectively based on their requirements.
Conclusion
Advancements in algorithms and software solution tools, such as deep learning and
pattern recognition, are driving the demand for faster processing chips and systems that
need to be integrated into most electronic devices that use AI. GTI designed its AI
accelerator chip for the seamless integration into electronic devices, improved energy
efficiency, lower production costs, and improved ease of use.
GTI’s success establishes a benchmark for companies planning to venture into the AI chip
market, which will be a major contributor to the growth of smart devices in the near
future. GTI expects to stay ahead in the competitive AI accelerator chip industry through
its innovations and customer-centric upgradations, thus earning the trust and accolades
from its clients.
For its strong overall performance, Gyrfalcon Technology Inc., has earned Frost &
Sullivan’s 2019 New Product Innovation Award in the North American AI accelerator chips
industry.
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 8 “We Accelerate Growth”
Significance of New Product Innovation
Ultimately, growth in any organization depends on continually introducing new products to
the market and successfully commercializing those products. For these dual goals to
occur, a company must be best in class in three key areas: understanding demand,
nurturing the brand, and differentiating from the competition.
Understanding New Product Innovation
Innovation is about finding a productive outlet for creativity—for consistently translating
ideas into high-quality products that have a profound impact on the customer.
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 9 “We Accelerate Growth”
Key Benchmarking Criteria
For the New Product Innovation Award, Frost & Sullivan analysts independently evaluated
2 key factors—New Product Attributes and Customer Impact—according to the criteria
identified below.
New Product Attributes
Criterion 1: Match to Needs
Criterion 2: Reliability
Criterion 3: Quality
Criterion 4: Positioning
Criterion 5: Design
Customer Impact
Criterion 1: Price/Performance Value
Criterion 2: Customer Purchase Experience
Criterion 3: Customer Ownership Experience
Criterion 4: Customer Service Experience
Criterion 5: Brand Equity
Best Practices Award Analysis for Gyrfalcon Technology
Decision Support Scorecard
To support its evaluation of best practices across multiple business performance categories, Frost & Sullivan employs a customized Decision Support Scorecard. This tool allows research and consulting teams to objectively analyze performance according to the key benchmarking criteria listed in the previous section, and to assign ratings on that
basis. The tool follows a 10-point scale that allows for nuances in performance evaluation. Ratings guidelines are illustrated below.
RATINGS GUIDELINES
The Decision Support Scorecard considers New Product Attributes and Customer Impact
(i.e., the overarching categories for all 10 benchmarking criteria; the definitions for each
criterion are provided beneath the scorecard). The research team confirms the veracity of
this weighted scorecard through sensitivity analysis, which confirms that small changes to
the ratings for a specific criterion do not lead to a significant change in the overall relative
rankings of the companies.
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 10 “We Accelerate Growth”
The results of this analysis are shown below. To remain unbiased and to protect the
interests of all organizations reviewed, Frost & Sullivan has chosen to refer to the other
key participants as Competitor 1 and Competitor 2.
Measurement of 1–10 (1 = poor; 10 = excellent)
New Product Innovation New Product Attributes
Customer Impact
Average
Rating
Gyrfalcon Technology Inc 9 9 9
Competitor 1 8 8 8
Competitor 2 7 7 7
New Product Attributes
Criterion 1: Match to Needs
Requirement: Customer needs directly influence and inspire the product’s design and
positioning.
Criterion 2: Reliability
Requirement: The product consistently meets or exceeds customer expectations for
consistent performance during its entire life cycle.
Criterion 3: Quality
Requirement: Product offers best-in-class quality, with a full complement of features and
functionalities.
Criterion 4: Positioning
Requirement: The product serves a unique, unmet need that competitors cannot easily
replicate.
Criterion 5: Design
Requirement: The product features an innovative design, enhancing both visual appeal
and ease of use.
Customer Impact
Criterion 1: Price/Performance Value
Requirement: Products or services offer the best value for the price, compared to similar
offerings in the market.
Criterion 2: Customer Purchase Experience
Requirement: Customers feel they are buying the optimal solution that addresses both
their unique needs and their unique constraints.
Criterion 3: Customer Ownership Experience
Requirement: Customers are proud to own the company’s product or service and have a
positive experience throughout the life of the product or service.
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 11 “We Accelerate Growth”
Criterion 4: Customer Service Experience
Requirement: Customer service is accessible, fast, stress-free, and of high quality.
Criterion 5: Brand Equity
Requirement: Customers have a positive view of the brand and exhibit high brand loyalty.
Decision Support Matrix
Once all companies have been evaluated according to the Decision Support Scorecard,
analysts then position the candidates on the matrix shown below, enabling them to
visualize which companies are truly breakthrough and which ones are not yet operating at
best-in-class levels.
High
Low
Low High
Cu
sto
mer I
mp
act
New Product Attributes
Gyrfalcon Technology Inc
Competitor 1
Competitor 2
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 12 “We Accelerate Growth”
Best Practices Recognition: 10 Steps to Researching, Identifying, and Recognizing Best Practices
Frost & Sullivan analysts follow a 10-step process to evaluate award candidates and
assess their fit with select best practices criteria. The reputation and integrity of the
awards are based on close adherence to this process.
STEP OBJECTIVE KEY ACTIVITIES OUTPUT
1 Monitor, target, and
screen
Identify award recipient candidates from around the
world
Conduct in-depth industry research
Identify emerging industries Scan multiple regions
Pipeline of candidates that potentially meet all best
practices criteria
2 Perform 360-degree
research
Perform comprehensive, 360-degree research on all
candidates in the pipeline
Interview thought leaders and industry practitioners
Assess candidates’ fit with best practices criteria
Rank all candidates
Matrix positioning of all candidates’ performance
relative to one another
3
Invite thought
leadership in best
practices
Perform in-depth examination of all candidates
Confirm best practices criteria
Examine eligibility of all candidates
Identify any information gaps
Detailed profiles of all ranked candidates
4
Initiate research
director review
Conduct an unbiased evaluation of all candidate
profiles
Brainstorm ranking options Invite multiple perspectives
on candidates’ performance Update candidate profiles
Final prioritization of all eligible candidates and
companion best practices positioning paper
5
Assemble
panel of industry
experts
Present findings to an expert
panel of industry thought leaders
Share findings
Strengthen cases for candidate eligibility
Prioritize candidates
Refined list of prioritized
award candidates
6
Conduct global
industry review
Build consensus on award candidates’ eligibility
Hold global team meeting to review all candidates
Pressure-test fit with criteria Confirm inclusion of all
eligible candidates
Final list of eligible award candidates, representing
success stories worldwide
7 Perform
quality check
Develop official award
consideration materials
Perform final performance
benchmarking activities Write nominations
Perform quality review
High-quality, accurate, and
creative presentation of nominees’ successes
8
Reconnect
with panel of industry
experts
Finalize the selection of the
best practices award recipient
Review analysis with panel
Build consensus Select recipient
Decision on which company
performs best against all best practices criteria
9 Communicate
recognition
Inform award recipient of recognition
Announce award to the CEO Inspire the organization for
continued success Celebrate the recipient’s
performance
Announcement of award and plan for how recipient
can use the award to enhance the brand
10 Take
strategic action
Upon licensing, company is
able to share award news with stakeholders and
customers
Coordinate media outreach
Design a marketing plan Assess award’s role in
strategic planning
Widespread awareness of
recipient’s award status among investors, media
personnel, and employees
BEST PRACTICES RESEARCH
© Frost & Sullivan 2019 13 “We Accelerate Growth”
The Intersection between 360-Degree Research and Best Practices Awards
Research Methodology
Frost & Sullivan’s 360-degree research
methodology represents the analytical
rigor of the research process. It offers a
360-degree view of industry challenges,
trends, and issues by integrating all 7 of
Frost & Sullivan’s research methodologies.
Too often companies make important
growth decisions based on a narrow
understanding of their environment,
resulting in errors of both omission and
commission. Successful growth strategies
are founded on a thorough understanding
of market, technical, economic, financial,
customer, best practices, and demographic
analyses. The integration of these research
disciplines into the 360-degree research
methodology provides an evaluation
platform for benchmarking industry
participants and for identifying those performing at best-in-class levels.
About Frost & Sullivan
Frost & Sullivan, the Growth Partnership Company, helps clients accelerate growth and
achieve best-in-class positions in growth, innovation, and leadership. The company's
Growth Partnership Service provides the CEO and the CEO's growth team with disciplined
research and best practices models to drive the generation, evaluation, and
implementation of powerful growth strategies. Frost & Sullivan leverages nearly 60 years
of experience in partnering with Global 1000 companies, emerging businesses, and the
investment community from 45 offices on 6 continents. To join Frost & Sullivan’s Growth
Partnership, visit http://www.frost.com.
360-DEGREE RESEARCH: SEEING ORDER IN THE CHAOS