<p>Tokyo Japan - LeapMind Inc., a creator of the standard in edge AI (Shibuya-ku, Tokyo; CEO: Soichi Matsuda) today announced the beta release of ultra-low power AI inference accelerator IP “Efficiera” version 2 (v2) before its commercial launch by the end of this year. Efficiera originally developed and licensed by LeapMind Inc. has been highly valued since its launch in October 2020 for its features of power saving, high performance, space saving, and performance scalability which enable easy installation to small FPGA and contribute to shortened development period of end users’ final products by supporting mass-produced boards. Upon the release of v2 beta version, LeapMind is welcoming trial use by and feedback from many people including SoC vendors and end user product designers. To obtain v2 beta version, please contact us at <a href="mailto:business@leapmind.io">business@leapmind.io</a></p>
<h2 id="main-specificationsfeatures-of-efficiera-v2">Main specifications/features of Efficiera v2</h2>
<ul>
<li>Great improvement of hardware performance: Achievement of 100TOPS/W from 27TOPS/W and speeding up of functions such as Skip connection, Convolution and Pixel Embedding</li>
<li>Flexible system development utilizing ASIC/ASSP as well as FPGA</li>
<li>Provision of extremely low bit quantization model development environment, “Efficiera NDK (Network Development Kit)” to meet the demands of end users who would like to develop individual models on Efficiera IP</li>
</ul>
<p><img src="/images/posts/en/news/7414/e62f6eaf-concept-660x371.png" alt=""></p>
<p>“It is a pleasure for us to release beta version while the development of Efficiera v2 is proceeding as scheduled”, commented Katsutoshi Yamazaki, LeapMind’s VP of Business. “Efficiera v2 is a product with a view to the use for ASIC/ASSP as well as FPGA, and its hardware performance enables image processing at the edge by utilizing machine learning. In the market, there is discussion on which is better, cloud processing or edge processing for complicated inference processing that requires large amounts of data, considering various conditions and constraints such as network load, power efficiency and real-time performance. Through our core technology, extremely low bit quantization which has developed under the assumption of inference processing using large volumes of complicated data on edge devices, we aim to help promote device development that allows users to choose inference processing on edge without hesitation. Efficiera v2 has achieved dramatically improved processing capability, especially for image processing, and our development department has confirmed that Efficiera v2 has demonstrated its great capability in situations where high-definition image processing by AI on edge devices is required. We hope that Efficiera v2 will trigger significant expansion of the conventional application scope of machine learning on edge”.</p>
<h2 id="about-efficiera">About Efficiera</h2>
<p>Efficiera is an ultra-low power AI inference accelerator IP specialized for CNN inference processing that runs as a circuit on FPGA or ASIC devices. The "ultra-small quantization" technology minimizes the number of quantization bits to 1 - 2 bits, maximizing the power and area efficiency of convolution, which accounts for most of the inference processing, without the need for advanced semiconductor manufacturing processes or special cell libraries. By using this product, deep learning functions can be incorporated into a variety of edge devices, including consumer electronics such as home appliances, industrial equipment such as construction machinery, surveillance cameras, broadcasting equipment, as well as small machines and robots that are constrained by power, cost, and heat dissipation, which has been technically difficult in the past. Visit product website at <a href="https://leapmind.io/products/efficiera-ip/">https://leapmind.io/products/efficiera-ip/</a></p>
<h2 id="about-leapmind">About LeapMind</h2>
<p>LeapMind Inc. was founded in 2012 with the corporate philosophy of "bringing new devices that use machine learning to the world". Total investment in LeapMind to date has reached 4.99 billion yen (as of May 2021). The company's strength is in extremely low bit quantization for compact deep learning solutions. It has a proven track record of achievement with over 150 companies, centered in manufacturing including the automobile industry. It is also developing its Efficiera semiconductor IP, based on its experience in the development of both software and hardware.</p>
<p>Head office: Shibuya Dogenzaka Sky Building 5F, 28-1 Maruyama-cho, Shibuya-ku, Tokyo 150-0044
Representative: Soichi Matsuda, CEO
Established: December 2012
URL: <a href="https://leapmind.io/en/">https://leapmind.io/en/</a>
LeapMind, Efficiera and logo are trademarks or registered trademarks of LeapMind Inc.
The information in the press release is current as of the date of announcement.</p>
<p>Media Contact
Efficiera Division, Marketing and Communication Team, LeapMind Inc.
Phone: 813-6696-6267
Email: <a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>

“Beta Version of Efficiera Version 2,” Ultra-low Power AI Inference Accelerator IP

<p>Hello and welcome to my LeapMind Summer Intern summary post!</p>
<h2 id="about-myself">About Myself</h2>
<p>I'm a German-born, Spanish-raised computer scientist with a passion for Neural Networks and Natural Language Processing. I'm a master student at the University of Tokyo and got the opportunity to join LeapMind for a summer itern program. Feel free to follow me on <a href="https://github.com/mariomeissner">GitHub</a> or <a href="https://twitter.com/mario_meissner">Twitter</a>! I also have my own <a href="https://mariomeissner.github.io/">blog</a> where I (very) occasionally write some stuff.</p>
<p>Let's start by giving some background about the topic at hand before I show you the results I obtained!</p>
<h2 id="the-lottery-ticket-hypothesis">The Lottery Ticket Hypothesis</h2>
<p><img src="/images/posts/en/news/summer-internship-2021-mario-finding-lottery-tickets-via-gradient-descent/b7f71fe6--2021-09-24-17.50.45-660x350.png" alt="Image: The Story of the Lottery Ticket Hypothesis: Hidden Networks"></p>
<p>Model pruning is a method to reduce the number of active weights in a neural network, for the purpose of decreasing model size and induce stronger regularization. A recent research direction called <a href="https://arxiv.org/abs/1803.03635">The Lottery Ticket</a> Hypothesis talks about how it is possible to find subnetworks that perform equally well (or even better!) than their full-sized counterparts, while also converging faster. This idea has prompted a number of follow-up works on pruning and lottery tickets and their relationship with the nature of how neural networks learn.</p>
<h2 id="finding-supermasks">Finding Supermasks</h2>
<p>One of the most notorious findings that spawned from the concept of Lottery Tickets is the supermask. This mightly creature is capable of pruning a randomly initialized model in such a way that it achieves full-sized model performance, but without ever changing the weights. Let that sink in for a moment! You randomly initialize the weights of a model, and then apply a certain mask, which zeroes-out a large amount of the weights. Without any weight training, this model performs extremely well on the test set of a certain task.</p>
<p>So what is this black magic? How do we obtain these supermasks? The answer is: gradient descent! Instead of calculating the gradients with respect to the weights, we compute the gradient on the mask and take a step towards a mask that reduces the target loss. Although applying a mask sounds like a non-differentiable task, via some tricks this is a totally possible. The algorithm that trains these masks is called edge-popup. Refer to the <a href="https://arxiv.org/abs/1911.13299">paper</a> and <a href="https://github.com/allenai/hidden-networks">its code</a> for details!</p>
<h2 id="supermasks-for-binarized-models">Supermasks for Binarized Models</h2>
<p>The <a href="https://arxiv.org/abs/2103.09377">Multi-Prize Lottery Ticket Hypothesis</a> paper talks about how we can find very robust binarized models via the supermask method. Models with binarized weights and activations are notoriously difficult to train, and here at LeapMind we wished to investigate the potential of this method to help us improve our own models.</p>
<h2 id="what-i-discovered">What I Discovered</h2>
<p><img src="/images/posts/en/news/summer-internship-2021-mario-finding-lottery-tickets-via-gradient-descent/6b400e0a--2021-09-24-8.36.36-660x350.png" alt="Image: Reproducing their results"></p>
<p>Just like in the Multi-Prize Lottery paper, a fully binarized lottery model remains 10%~20% below in accuracy against a standard weight-trained float model, especially for larger datasets such as ImageNet. However, comparing a binarized lottery model against a weight-trained binarized model is a different story! Here we can see extremely competitive results depending on the model configuration and dataset.</p>
<p>However, an important realization can be made to further improve results. The idea of freezing the weights and performing only mask-optimization is interesting from a theoretical point of view and helps these research papers get published, but when it comes down to the practical benefit, there is no reason for us to restrain ourselves in such a way! Why not combining the power of supermasks with the old-fashioned but well-known weight-training?</p>
<p>It turns out that supermask training in combination with normal weight-training acts like a strong regularizer, helping to improve generalization and avoid learning spurious correlations. Overall, we obtain much better training curves when training an unfrozen model with supermask training.</p>
<p>Additionally, assuming the existence of a powerful sparse-multiplication library, we can speed-up inference time of these models due to the fact that the weights are masked and thus very sparse.</p>
<h2 id="conclusions">Conclusions</h2>
<p>This summer internship was extremely exciting and I learned a ton of things. First and foremost, I got to dive deep into the theoretical guts of model pruning, lottery tickets and edge-popup supermask training. An additional, and in my opinion even more valuable experience was to see firsthand how a large, cutting-edge deep learning repository is organized and maintained. I learned how to <a href="https://techblog.leapmind.io/blog/boost-your-ai-research-with-hydra-and-lightning-templates/">structure complex research experiments</a> aiming for clarity and reproducibility. This will be my biggest takeaway for sure!</p>

Summer Internship 2021 – Mario – Finding Lottery Tickets via Gradient Descent!

<p>June 22, 2021 - LeapMind Inc. (Shibuya-ku, Tokyo; CEO: Soichi Matsuda), a creator of the standard in edge AI today announced that it has been granted two patents for its extremely low bit quantization technology. “Two patents we acquired are one of the fundamental technologies supporting Efficiera, an ultra-low power AI inference accelerator IP that we developed and provided.” Commented Hiroyuki Tokunaga, CTO of LeapMind and one of the inventors of patents acquired. “We continue to develop related technologies and file patent applications.”</p>
<p><strong>Patent number: No. 6885645</strong>
<strong>Date of patent registration:</strong> May 17, 2021
<strong>Title of the invention:</strong> Neural network processing units, neural network processing method, and neural network processing program, neural network processing program
<strong>Patentee:</strong> LeapMind Inc.</p>
<h2 id="patent-summary">Patent Summary</h2>
<p>This patent relates to a technology that minimizes the accuracy degradation associated with low bit rates, which is an issue in extremely low bit quantization technology. This is one of the methods of quantizing multi-bit data such as images, in which the multi-bit elements in the data are replaced by quantized vectors instead of simply being reduced to lower bits. In particular, the use of this technology in the input layer, which is the first stage of a deep neural network, can be highly effective.</p>
<p><strong>Patent number: No. 6886747</strong>
<strong>Patent registration date:</strong> May 19, 2021
<strong>Title of the invention:</strong> Neural network processing units, neural network processing method, and neural network processing program, neural network processing program
<strong>Patentee:</strong> LeapMind Inc.</p>
<h2 id="patent-summary-1">Patent Summary</h2>
<p>The quantization operation is a comparison between the input value and the threshold value. Therefore, arbitrary monotonic functions between convolution and quantization operations can eventually be converted into threshold changes, which can be performed efficiently.</p>
<h2 id="about-leapmind">About LeapMind</h2>
<p>LeapMind Inc. was founded in 2012 with the corporate philosophy of "bringing new devices that use machine learning to the world". Total investment in LeapMind to date has reached 4.99 billion yen (as of May 2021). The company's strength is in extremely low bit quantization for compact deep learning solutions. It has a proven track record of achievement with over 150 companies, centered in manufacturing including the automobile industry. It is also developing its Efficiera semiconductor IP, based on its experience in the development of both software and hardware.</p>
<p>Head office: Shibuya Dogenzaka Sky Building 5F, 28-1 Maruyama-cho, Shibuya-ku, Tokyo 150-0044
Representative: Soichi Matsuda, CEO
Established: December 2012
URL：<a href="https://leapmind.io/en/">https://leapmind.io/en/</a></p>
<p>LeapMind, Efficiera and logo are trademarks or regidtered trademarks of LeapMind Inc.
The infomation in the press release is current as of the date of announcement.</p>
<p><strong>Media Contact</strong>
Marketing Communication Division, Public Relations Officer, LeapMind Inc.
Phone: 813-6696-6267
Email: <a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>

【Press Release】LeapMind Acquires Patents for Extremely Low Bit Quantization Technology

<p>May 20th, 2021, Tokyo Japan - LeapMind Inc., a creator of the standard in edge AI (Shibuya-ku, Tokyo; CEO: Soichi Matsuda) today announced that company’s proprietary ultra-low power AI inference accelerator IP “Efficiera” was verified RTL design for ASIC/ASSP.</p>
<p>“By conducting the design verification this time, we were able to confirm the expected PPA (Power/Performance/Area) at the time of IP configuration.” Said Katsutoshi Yamazaki, VP of Business at LeapMind. “This is a big step for us moving forward to future LSI commercialization of Efficiera”.</p>
<p>Efficiera is an ultra-low power consumption AI inference accelerator IP specialized for CNN inference arithmetic processing that operates as a circuit on FPGA devices or ASIC/ASSP devices. For more information, visit here (<a href="https://leapmind.io/products/efficiera-ip/">https://leapmind.io/products/efficiera-ip/</a>).</p>
<h2 id="about-leapmind">About LeapMind</h2>
<p>LeapMind Inc. was founded in 2012 with the corporate philosophy of "bringing new devices that use machine learning to the world". Total investment in LeapMind to date has reached 4.99 billion yen (as of May 2021). The company's strength is in extremely low bit quantization for compact deep learning solutions. It has a proven track record of achievement with over 150 companies, centered in manufacturing including the automobile industry. It is also developing its Efficiera semiconductor IP, based on its experience in the development of both software and hardware.</p>
<p>Head office: Shibuya Dogenzaka Sky Building 5F, 28-1 Maruyama-cho, Shibuya-ku, Tokyo 150-0044
Representative: Soichi Matsuda, CEO
Established: December 2012
URL：<a href="https://leapmind.io/en/">https://leapmind.io/en/</a></p>
<p>*LeapMind, Efficiera and logo are trademarks or registered trademarks of LeapMind Inc.
*The information in the press release is current as of the date of announcement.</p>
<p><strong>Media Contact</strong>
Marketing Communication Division, Public Relations Officer, LeapMind Inc.
TEL: 813-6696-6267
Email: <a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>
<p><strong>Inquiries about Products and Services</strong>
Efficiera Division, Business Development Team, LeapMind Inc.
TEL: 813-6696-6267
Email: <a href="mailto:business@leapmind.io">business@leapmind.io</a>
Product official website URL: <a href="https://leapmind.io/products/efficiera-ip/">https://leapmind.io/products/efficiera-ip/</a></p>

【Press Release】LeapMind’s “Efficiera” Ultra-low Power AI Inference Accelerator IP Was Verified RTL Design for ASIC/ASSP Conversion

<h2 id="eizo-corporation-completes-joint-performance-evaluation-of-ultra-low-power-ai-inference-accelerator-ip-efficiera">EIZO Corporation Completes Joint Performance Evaluation of Ultra Low Power AI Inference Accelerator IP "Efficiera"</h2>
<p>Completed joint performance evaluation of Efficiera with EIZO's visibility-enhancing system, for construction of an edge AI system</p>
<p>February 4th, 2021 – Tokyo, Japan - LeapMind Inc. (Shibuya, Tokyo; CEO: Soichi Matsuda), creator of the standard in edge AI, and EIZO Corporation (Hakusan, Ishikawa; President and CEO: Yoshitaka Jitsumori) have completed a joint performance evaluation of LeapMind's Efficiera ultra low power AI inference accelerator IP using EIZO's DuraVision EVS1VX visibility-enhancing system (released in Japan on August 25, 2020), for the construction of an edge AI system. The two companies will continue to aim for the goal of solving social issues by promoting joint development and evaluation utilizing Efficiera and achieving the installation of edge AI functions in video equipment.</p>
<p><img src="/images/posts/en/news/7119/bae09cbe-hp_en.png" alt="Images of Efficiera and DuraVision-EVS1VX" title="Images of Efficiera and DuraVision-EVS1VX"></p>
<h2 id="overview-of-joint-performance-evaluation">Overview of Joint Performance Evaluation</h2>
<p>EIZO collaborated in this joint performance evaluation based on their belief that the Efficiera ultra low power AI inference accelerator IP, which is a commercialized form of LeapMind's extremely low bit quantization, could be used in the construction of an edge AI system that will add analysis functions to EIZO's DuraVision EVS1VX visibility-enhancing system.
By taking one of the features of Efficiera, the small footprint, it is possible to utilize the free space of the FPGA that is already installed in the DuraVision EVS1VX to achieve deep learning functions, without adding to or changing the hardware.
This provides the advantage of keeping the overall system cost low.
Through this joint performance evaluation, EIZO confirmed that it can achieve practical performance, and assessed the efficacy of Efficiera.
Verification of the object detection model was completed in this joint performance evaluation. The companies
will proceed with the verification of new models that will be added moving forward.
LeapMind and EIZO will continue to use Efficiera to promote joint development and evaluation, with the goal of putting edge AI systems to practical application.</p>
<h2 id="comment-from-eizo-corporation">Comment from EIZO Corporation</h2>
<p>As a visual technology company, EIZO works to provide imaging solutions to address issues faced by society and our customers. In collaboration with LeapMind, we have conducted a performance evaluation of Efficiera in the DuraVision EVS1VX visibility-enhancing system that was released on August 25, 2020. We view this as a promising technology for quickly adding analysis functions to high-quality and high-functionality video equipment that will allow us to make flexible proposals to address the diverse and complicated issues of today.
Following four evaluations were conducted and the results confirmed the features of Efficiera and its usefulness:</p>
<p>(1) Resource usage rate on FPGA
(2) Inference processing speed on FPGA
(3) Accuracy of object detection
(4) Generalization of object detection model</p>
<p>Utilizing technical support based on LeapMind's extensive knowledge, including tuning methods and the number of datasets, we were able to proceed with development and evaluation quickly, and were able to obtain a satisfactory level of accuracy in a short period of time.
In the future, EIZO will use verification testing to further improve accuracy, in order to utilize this technology to solve social issues.</p>
<h2 id="comment-from-leapmind-inc">Comment from LeapMind Inc.</h2>
<p>EIZO has had its eye on LeapMind's technology even before the commercialization of Efficiera, and they conducted their evaluation as this product was brought to market.</p>
<ul>
<li>We believe that this evaluation has confirmed the goals that we aimed for with Efficiera:</li>
<li>By installing Efficiera in the free space of an existing FPGA, deep learning inference functions can be added without additional BoM cost.</li>
<li>Circuit integration is possible in the actual FPGA design, not just the reference environment, achieving the target performance for challenges including object detection models.</li>
</ul>
<p>Fine tuning tools allow custom object detection models that are optimized for Efficiera to be developed without coding.
The feedback that we received during this evaluation will be applied to Efficiera products as soon as possible, and we will continue to make improvements in order to provide even better products and services. By continuing our efforts with EIZO, LeapMind will accelerate the practical application of edge AI systems that utilize Efficiera.</p>
<h2 id="about-eizos-duravision-evs1vx">About EIZO's DuraVision EVS1VX</h2>
<p>DuraVision EVS1VX analyzes video images to determine which data requires correction, performs visibility enhancement processing to make the data easier to recognize, and then displays the corrected data on the connected monitor in real time. This makes it easier to detect abnormalities visually and improves the accuracy of detecting abnormalities with AI. Therefore, this technology is expected to support the prevention of crime, the reliable detection of defects, and the safe maintenance of equipment, by allowing stable operation without relying on the skill of an operator.
Official website (Japanese): <a href="https://www.eizo.co.jp/products/id/evs1vx/">https://www.eizo.co.jp/products/id/evs1vx/</a></p>
<p><img src="/images/posts/en/news/7119/34486b5a--0003-02-04-1.29.22.png" alt="Description of steady-state monitoring of tunnels." title="Description of steady-state monitoring of tunnels."></p>
<h2 id="about-leapminds-efficiera">About LeapMind's Efficiera</h2>
<p>Efficiera is an ultra-low power AI inference accelerator IP that operates as a circuit on FPGA or ASIC/ASSP devices, and is specialized for CNN inference operations. It is scheduled to be officially released this autumn. Since the power efficiency and area efficiency of the convolution (which account for most of the inference processing) are maximized by the extremely low bit quantization that minimizes quantization to 1 or 2 bits, there is no need to use state-of-the-art semiconductor manufacturing processes or special cell libraries.
It allows deep learning capabilities to be embedded in home electronics and other consumer appliances, construction machinery and other industrial equipment, surveillance cameras, and broadcasting equipment. It can also be embedded in various edge devices where it was technically difficult to add deep learning capabilities before due to restrictions on power, cost, and heat dissipation, such as small machines and robots.
Official website：<a href="https://leapmind.io/products/efficiera-ip/">https://leapmind.io/products/efficiera-ip/</a></p>
<h2 id="about-eizo-corporation">About EIZO Corporation</h2>
<p>EIZO (TSE:6737), which means image in Japanese, is a visual technology company that develops and manufactures high-end visual solutions to enrich people in their professional and personal lives. EIZO offers total imaging solutions with a range of monitors, software, video capture, processing, and distribution solutions, cameras, and advanced integrated technologies to meet the specialized needs of customers in business, creative fields, healthcare, air traffic control, maritime, security &#x26; surveillance, and more. Headquartered in Hakusan, Japan, EIZO has R&#x26;D and manufacturing facilities in Japan, China, Germany, and the United States, and representation in more than 90 countries.</p>
<h2 id="about-leapmind-inc">About LeapMind Inc.</h2>
<p>LeapMind Inc. was founded in 2012 with the corporate philosophy of "bringing new devices that use machine learning to the world". Total investment in LeapMind to date has reached 4.99 billion yen. The company's strength is in extremely low bit quantization for compact deep learning solutions. It has a proven track record of achievement with over 150 companies, centered in manufacturing including the automobile industry. It is also developing its Efficiera semiconductor IP, based on its experience in the development of both software and hardware.</p>
<p>Head office: Shibuya Dogenzaka Sky Building 5F, 28-1 Maruyama-cho, Shibuya-ku, Tokyo 150-0044
Representative: Soichi Matsuda, CEO
Established: December 2012
URL：<a href="https://leapmind.io/en/">https://leapmind.io/en/</a>
*EIZO and DuraVision are registered trademarks of EIZO Corporation in Japan and other countries. Other product names in this text are trademarks or registered trademarks of LeapMind Inc.
*The information in this press release is current as of the date of publication.</p>
<p><strong>Media Contact</strong>
Shiho Takai, Marketing Communication Division, LeapMind Inc.
TEL: 813-6696-6267, EMAIL: <a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>
<p><strong>Sales Contact</strong>
Efficiera Division, Business Development Team, LeapMind Inc.
TEL：813-6696-6267, EMAIL：<a href="mailto:business@leapmind.io">business@leapmind.io</a>
LeapMind株式会社
Blueoil Division　Business Development Team
TEL：03-6696-6267
MAIL：<a href="mailto:business@leapmind.io">business@leapmind.io</a></p>

【Press Release】EIZO Corporation Completes Joint Performance Evaluation of Ultra Low Power AI Inference Accelerator IP “Efficiera”

<h2 id="leapmind-announces-participation-in-alliance-program-of-xilinx-a-major-us-fpga-innovator">LeapMind Announces Participation in Alliance Program of Xilinx, a Major US FPGA Innovator</h2>
<p>Accelerating the Practical Application of Edge AI Through Efficiera Ultra-Low Power AI Inference Accelerator IP Core Support for Xilinx FPGAs</p>
<p>LeapMind, Inc. (Location: Shibuya, Tokyo; CEO: Soichi Matsuda; hereinafter referred to as "LeapMind"), which makes the standard of Edge AI (Artificial Intelligence), announces its participation in the Xilinx, Inc. Alliance Program (Head Office: San Jose, California, USA / NASDAQ: XLNX). In the future, LeapMind will provide customers with lower cost, faster FPGA solutions by linking the Efficiera ultra-low power AI inference accelerator IP core with platforms from Xilinx, an FPGA innovator.</p>
<h2 id="outline">Outline</h2>
<p>The Xilinx Alliance Program is a global ecosystem of qualified companies that are engaged in acceleration solutions, IP cores, design services, and board development and production. By participating in this alliance program, LeapMind, which develops the Efficiera semiconductor IP core that runs deep learning on FPGAs, will be able to meet the challenges of a wide range of customers by strengthening its collaboration with Xilinx, a leader in the global FPGA market, expanding its business through joint marketing activities, and developing solutions globally. In collaboration with Xilinx, LeapMind will provide solutions that are highly competitive in the market to further accelerate the practical application of edge AI utilizing FPGAs.</p>
<h2 id="endorsement-by-xilinx-inc">Endorsement by Xilinx, Inc.</h2>
<p>"Thank you for making LeapMind's AI solutions compatible with Xilinx boards. 　Through the Xilinx Alliance Program, we support everything from product concepts to mass production, and provide a ready-to-use hardware platform to shorten development time and increase productivity. We also provide a variety of services in all aspects of product design and manufacturing , and possess the know-how and knowledge for shortening product development cycles and reducing risk. 　We believe that LeapMind's participation in this program will allow more edge AI products that utilize FPGAs to be released to the world. We are looking forward to Efficiera."</p>
<p>Hiroshi Hayashida
Country Manager
Xilinx, Inc.</p>
<h2 id="about-xilinx-inc-and-the-xilinx-alliance-program">About Xilinx, Inc. and the Xilinx Alliance Program</h2>
<p>ilinx, Inc. develops highly flexible and adaptive processing platforms that enable rapid innovation across a variety of technologies - from the cloud, to the edge, to the endpoint. Xilinx is the inventor of the FPGA and Adaptive SoCs (including our Adaptive Compute Acceleration Platform, or ACAP), designed to deliver the most dynamic computing technology in the industry. We collaborate with our customers to create scalable, differentiated and intelligent solutions that enable the adaptable, intelligent and connected world of the future.For more information, visit <a href="https://www.amd.com/en.html">xilinx.com</a></p>
<h2 id="about-efficiera">About Efficiera</h2>
<p>Efficiera is an ultra-low power AI inference accelerator IP core that operates as a circuit on FPGA or ASIC devices, and is specialized for CNN inference operations. It is scheduled to be officially released this autumn. Since the power efficiency and area efficiency of the convolution (which account for most of the inference processing) are maximized by the extremely low bit quantization technology that minimizes quantization to 1 or 2 bits, there is no need to use state-of-the-art semiconductor manufacturing processes or special cell libraries. It allows deep learning capabilities to be embedded in home electronics and other consumer appliances, construction machinery and other industrial equipment, surveillance cameras, and broadcasting equipment. It can also be embedded in various edge devices where it was technically difficult to add deep learning capabilities before due to restrictions on power, cost, and heat dissipation, such as small machines and robots.
Official website：<a href="https://leapmind.io/products/efficiera-ip/">https://leapmind.io/products/efficiera-ip/</a></p>
<h2 id="about-leapmind-inc">About LeapMind, Inc.</h2>
<p>LeapMind, Inc. was founded in 2012 with the corporate philosophy of "bringing new devices that use machine learning to the world". Total investment in LeapMind to date has reached 4.99 billion yen.
The company's strength is in extremely low bit quantization for compact deep learning solutions. It has a proven track record of achievement with over 150 companies, centered in manufacturing including the automobile industry. It is also developing its Efficiera semiconductor IP core, based on its experience in the development of both software and hardware.
Head office: Shibuya Dogenzaka Sky Building 5F, 28-1 Maruyama-cho, Shibuya-ku, Tokyo 150-0044
Representative: Soichi Matsuda, CEO
Established: December 2012
URL：<a href="https://leapmind.io/en/">https://leapmind.io/en/</a></p>
<p>*Product names in this text are trademarks or registered trademarks of LeapMind, Inc.
*The information in this press release is current as of the date of publication.</p>
<p><strong>Press Relations Contact Information</strong>
LeapMind, Inc.
Takai, Marketing Communication Division
Tel:+81-(0)3-6696-6267
E-mail：<a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>
<p>**Products Team Contact Information **
LeapMind, Inc.
Business Development Team, Efficiera Division
Tel:+81-(0)3-6696-6267
E-mail：<a href="mailto:business@leapmind.io">business@leapmind.io</a>
Official website：<a href="https://leapmind.io/products/efficiera-ip/">https://leapmind.io/products/efficiera-ip/</a></p>

【Press Release】LeapMind Announces Participation in Alliance Program of Xilinx, a Major US FPGA Innovator

<p>Left: Naohide Yabuki, President and CEO, PALTEK Corporation 　Right: Katsutoshi Yamazaki, VP &#x26; GM, Blueoil Division, LeapMind, Inc.</p>
<h2 id="efficiera-officially-supports-palteks-so-one-module-with-the-aim-of-providing-ai-solutions-for-fpgas-in-industrial-equipment">Efficiera officially supports PALTEK's So-One Module, with the aim of providing AI solutions for FPGAs in industrial equipment</h2>
<p>LeapMind, Inc. (Location: Shibuya, Tokyo; CEO: Soichi Matsuda; hereinafter referred to as "LeapMind"), which provides solutions for companies that utilize deep learning technology, an elemental technology of AI (Artificial Intelligence), has partnered with PALTEK Corporation (Head Office: Kohoku-ku, Yokohama; President: Naohide Yabuki; Securities code: 7587; hereinafter referred to as "PALTEK") on an edge AI solution utilizing FPGAs. This will allow LeapMind to work with PALTEK to jointly provide optimal AI solutions to customers, toward the practical application of edge AI solutions. In addition, LeapMind's Efficiera ultra-low power AI inference accelerator now officially supports the So-One Module launched by PALTEK. This will shorten development lead-time for mass-produced products equipped with deep learning capabilities, and enable the mass production of on-device AI solutions.</p>
<h2 id="background-and-outline-of-this-collaboration">Background and Outline of This Collaboration</h2>
<p>In recent years, dramatic improvements have been made to the accuracy of deep learning, an elemental technology of AI, in a variety of fields such as image processing and voice processing. At present, it is common for processing to be performed using GPUs in the cloud. However, there are many cases in which efficient processing is not possible in the cloud, due to problems of communication volume, load, and latency at the time of inference. Performing inference processing on edge devices has become an important technical issue.</p>
<p>In particular, real-time IoT applications such as autonomous driving, smart cities, and robot control, which have strict restrictions regarding response speed and network environment, require inference processing with low-end devices that use little power and memory. Attention has been focused on using FPGAs to solve problems in the area of edge AI. FPGAs can achieve high performance and low latency with low power compared to CPUs and GPUs, and have long-term warranties and durability.</p>
<p>Since its founding in 1982, PALTEK, a participant in LeapMind's Efficiera FPGA Partner Program, has used FPGAs to provide product development support and technical support to Japanese electronics manufacturers. The company's strength, cultivated through its semiconductor sales, lies in its ability to support customers from proposal through to mass production, to create original boards such as the So-One Module, and integrated design services from design specification through mass production.</p>
<p>Based on this background, a collaboration was formed combining LeapMind's advanced technology for deploying AI capabilities in small edge devices such as low power SoC FPGAs with PALTEK's wealth of experience in industrial equipment, proposals, sales capabilities, and technical support through the introduction of mass production, making it possible to provide a high-precision edge AI solution package that meets the needs of the market within a short time frame.</p>
<h2 id="main-initiativescollaboration-method">Main Initiatives/Collaboration Method</h2>
<p>With this collaboration, the strengths of both companies are combined to accelerate the development and mass production of edge AI solutions utilizing FPGAs. Both companies will propose and implement optimal solutions for solving customers' issues. Through joint sales expansion activities, solutions that integrate the So-One Module with Efficiera will be offered to select businesses initially. It will then become officially available to a large number of businesses that are considering the introduction of AI technology to industrial equipment such as construction and agricultural machinery.</p>
<p><strong>Naohide Yabuki, President of PALTEK Corporation, had the following to say about this collaboration:</strong>
"We are very pleased to sign the Efficiera FPGA Partner Program Agreement in order to be able to collaborate with LeapMind. Combining LeapMind's Efficiera with our So-One Module will open a wide range of new possibilities for our customers, such as the industrial equipment manufacturers with whom we have been working. We are confident that through this collaboration we will overcome any PoC issues and accelerate the practical application of mass-produced edge AI products."</p>
<p><strong>Katsutoshi Yamazaki, VP &#x26; GM of Blueoil Division, LeapMind, Inc., had the following to say about this collaboration:</strong>
"We are confident that our collaboration with PALTEK Corporation will allow us to accelerate the expansion of the edge AI business utilizing FPGAs. We feel very strongly that we will be able to provide integrated solutions together with PALTEK Corporation, which has a wealth of experience in industrial equipment. We will make proposals together to enhance the value of the customer experience and contribute to the practical application of our customers' edge AI products. LeapMind will continue to provide key technologies for the future, aiming to achieve our mission of bringing devices that use machine learning to the world."</p>
<h2 id="official-efficiera-support-for-so-one-module">Official Efficiera Support for So-One Module</h2>
<p>LeapMind's Efficiera AI inference accelerator IP core now officially supports the So-One Module original hardware platform for edge AI from PALTEK. This will shorten development lead-time and reduce the cost for mass-produced products with deep learning capabilities, and accelerate the mass production of on-device AI solutions for customers. LeapMind will continue to expand Efficiera support for mass-produced boards, including the So-One Module.</p>
<p><img src="/images/posts/en/news/7016/cea0dd9d--12.png" alt="Explanation of the official support of Efficiera for the So-One module." title="Explanation of the official support of Efficiera for the So-One module."></p>
<h2 id="efficiera-fpga-partner-program">Efficiera FPGA Partner Program</h2>
<p>The Efficiera FPGA Partner Program was launched with the goal of co-creation of on-device AI systems and solutions that solve customer issues. By participating in this program, partners combine their products and services with Efficiera in order to develop and provide package services and systems that meet their customer's needs, and develop package services that add value to Efficiera.</p>
<p>Please feel free to contact us at the following URL if you would like to participate in the Efficiera FPGA Partner Program.
▶ Efficiera FPGA Partner Program Contact Form:<a href="https://leapmind.io/en/contact/other/">https://leapmind.io/en/contact/other/</a></p>
<h2 id="so-one-module">So-One Module</h2>
<p>The embedded FPGA edge computing So-One Module developed by PALTEK and equipped with Xilinx's MPSoC* is the world's smallest MPSoC-equipped module capable of edge computing. The So-One Module enables mounting high-speed edge computing devices on drones, which has always proven a difficult challenge. In addition, installation in surveillance cameras in retail locations and line sensor cameras at manufacturing sites can greatly improve edge computing performance. This allows the acceleration of workloads such as AI-based lesion diagnosis support, sensor data analysis at factory facilities, and image analysis with industrial robots using vision sensors, leading to reduced system costs and power consumption.
For more information about the So-One Module from PALTEK, please visit <a href="https://www.paltek.co.jp/newsrelease/press201020_So-One_Module/index.html">https://www.paltek.co.jp/newsrelease/press201020_So-One_Module/index.html</a></p>
<p>*MPSoC MPSoC, part of Xilinx's SoC portfolio, combines processor software programmability with FPGA hardware programmability for exceptional system performance, flexibility, and scalability. It also brings great benefits to the entire system, such as the ability to quickly commercialize low-power, low-cost designs.</p>
<p><img src="/images/posts/en/news/7016/ff668873-img_8579-2.png" alt="Image of MPSoC alone and comparison between a 500 yen coin and MPSoC." title="Image of MPSoC alone and comparison between a 500 yen coin and MPSoC."></p>
<h2 id="efficiera">Efficiera</h2>
<p>Efficiera is an ultra-low power AI inference accelerator IP core that operates as a circuit on FPGA or ASIC devices, and is specialized for CNN inference operations. It is scheduled to be officially released this autumn. Since the power efficiency and area efficiency of the convolution (which account for most of the inference processing) are maximized by the extremely low bit quantization technology that minimizes quantization to 1 or 2 bits, there is no need to use state-of-the-art semiconductor manufacturing processes or special cell libraries. It allows deep learning capabilities to be embedded in home electronics and other consumer appliances, construction machinery and other industrial equipment, surveillance cameras, and broadcasting equipment. It can also be embedded in various edge devices where it was technically difficult to add deep learning capabilities before due to restrictions on power, cost, and heat dissipation, such as small machines and robots.
Official website:<a href="https://leapmind.io/en/products/efficiera-ip/">https://leapmind.io/en/products/efficiera-ip/</a></p>
<p>*Efficiera is a registered trademark of LeapMind, Inc. * The Xilinx name and Zynq, UltraScale +, and other brand names mentioned in this press release are registered trademarks or trademarks of Xilinx in the United States and other countries. All other names belong to their respective owners.</p>
<h2 id="about-paltek-corporation">About PALTEK Corporation</h2>
<p>Since its founding in 1982, PALTEK has provided semiconductor products in Japan and overseas to Japanese electronics manufacturers, as well as providing design contract services for hardware and software. As a partner in product development for customers, the company provides support from the examination of specifications to trial development and mass production. PALTEK contributes to our customers' development by providing optimal solutions based on our corporate philosophy of "coexistence with others".
For more information about PALTEK, please visit <a href="https://www.paltek.co.jp">https://www.paltek.co.jp</a></p>
<h2 id="about-leapmind-inc">About LeapMind, Inc.</h2>
<p>LeapMind, Inc. was founded in 2012 with the corporate philosophy of "bringing new devices that use machine learning to the world". Total investment in LeapMind to date has reached 4.99 billion yen. The company's strength is in extremely low bit quantization for compact deep learning solutions. It has a proven track record of achievement with over 150 companies, centered in manufacturing including the automobile industry. It is also developing its Efficiera semiconductor IP core, based on its experience in the development of both software and hardware.</p>
<p>Head office: Shibuya Dogenzaka Sky Building 5F, 28-1 Maruyama-cho, Shibuya-ku, Tokyo 150-0044
Representative: Soichi Matsuda, CEO
Established: December 2012
URL:<a href="https://leapmind.io/en/">https://leapmind.io/en/</a></p>
<p><strong>For further information on this matter, please contact.</strong>
LeapMind, Inc.
PR/Branding Division　PR manager
TEL：03-6696-6267
MAIL：<a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>
<p><strong>PALTEK Corporation.</strong>
Contact person ：State of the Art Solution　AI Product
MAIL ： <a href="mailto:info_pal@paltek.co.jp">info_pal@paltek.co.jp</a>
ADDRESS ： 6F Shin-Yokohama Square Building, 2-3-12 Shin-Yokohama, Kohoku-ku, Yokohama City, Kanagawa Prefecture
TEL ： 045-477-2001
FAX ： 045-477-2010</p>
<p><strong>Inquiries about Products and Services</strong>
LeapMind, Inc.
Business Development Team, Blueoil Division
Tel:03-6696-6267
E-mail:<a href="mailto:business@leapmind.io">business@leapmind.io</a>
Official website:<a href="https://leapmind.io/en/products/efficiera-ip/">https://leapmind.io/en/products/efficiera-ip/</a></p>

【Press Release】LeapMind Collaborates with PALTEK to Accelerate the Development and Mass Production of Edge AI Solutions Utilizing FPGAs

<h2 id="a-new-standard-for-edge-ai-with-the-worlds-first-hardware-and-software-platform-utilizing-extremely-low-bit-quantization">A New Standard for Edge AI with the World's First Hardware and Software Platform Utilizing Extremely Low Bit Quantization</h2>
<p>LeapMind, Inc. (Location: Shibuya, Tokyo; CEO: Soichi Matsuda; hereinafter referred to as "LeapMind"), which provides solutions for companies that utilize deep learning technology, an elemental technology of AI (Artificial Intelligence), is pleased to announce that it has officially launched the commercial version of its Efficiera ultra-low power consumption AI inference accelerator IP core, which incorporates deep learning capabilities into various edge devices.</p>
<h2 id="overview-of-commercial-version-of-efficiera">Overview of Commercial Version of Efficiera</h2>
<p>Efficiera is an ultra-low power AI inference accelerator IP core that is specialized for CNN (*1) inference operations. Deep learning capabilities can be incorporated into various edge devices including construction equipment and other industrial machinery, surveillance cameras, and broadcasting equipment. Efficiera has been described as an RTL for use with both ASIC and FPGA, but at the time of this release it has a semiconductor IP design suitable for FPGA circuits. It uses extremely low bit quantization (*2) to provide trained object detection AI models optimized for Efficiera. By using these models in combination, performance of Efficiera can be maximized.</p>
<p><img src="/images/posts/en/news/7011/3042865b--7.png" alt="Overview of the commercial version of Efficiera." title="Overview of the commercial version of Efficiera."></p>
<p>Figure 1: Provided Model and How It is Used</p>
<h2 id="features-of-efficiera">Features of Efficiera</h2>
<p>LeapMind has developed Efficiera based on knowledge gained from the co-creation of machine learning solutions with more than 150 companies, as well as through its research and development of technologies including extremely low bit quantization. It offers the following 4 features.</p>
<ul>
<li>Energy Efficient
By minimizing the volume of data transmitted and the number of bits, the power required for convolution operations is reduced.</li>
<li>High Performance
By reducing the arithmetic logic complexity, the number of operation cycles is reduced and the arithmetic capacity per area/clock rate is improved.</li>
<li>Small Footprint
By minimizing the number of operated bits, the circuit area and SRAM size per arithmetic logic unit are minimized.</li>
<li>Scalable
Since the computing performance can be fine-tuned by adjusting the circuit configuration, it is possible to optimize the configuration and maximize the performance of Efficiera according to the task being performed.</li>
</ul>
<p>We also offer the following software and functions to allow you to get the most out of Efficiera.</p>
<ul>
<li>Optimized, Trained AI Models
We provide trained models optimized for Efficiera.</li>
<li>Fine Tuning
We also provide tools for "fine tuning" – reusing part of an existing model to help train a new model for your specific needs. By leveraging the fine-tuning tool and existing models, it is possible to build new models with less data and without the trouble of creating a new model from scratch.</li>
</ul>
<h2 id="benefits-of-efficiera-with-fpga">Benefits of Efficiera with FPGA</h2>
<p><strong>Mount on small FPGAs</strong>
The small footprint of Efficiera allows it to be mounted on small FPGAs. In the minimum configuration, it can be mounted with about 1/3 of the ALM resources of Intel's Cyclone® V SoC with 110K LEs, and the performance can be doubled or quadrupled as the resources allow. Resource usage and theoretical performance at 125 MHz are shown in Figure 2.</p>
<p><img src="/images/posts/en/news/7011/a1b789e6--8.png" alt="Pie chart of resource usage vs. theoretical performance" title="Pie chart of resource usage vs. theoretical performance"></p>
<p>Figure 2: Resource Usage and Theoretical Performance</p>
<p><strong>Add AI capabilities without changing the hardware design</strong>
Since the IP core itself is small, Efficiera can be integrated on the same SoC FPGA device as the CPU and image input circuit. This allows a lower BoM (Bill Of Materials) for products equipped with AI capabilities. For example, by mounting Efficiera in the free space of an existing FPGA design, AI capabilities can be added without changing the number of components.</p>
<p><strong>Mass-produced boards to shorten development lead-time</strong>
LeapMind is promoting the Efficiera FPGA Partner Program, and is working to make Efficiera compatible with even more mass-produced boards as part of this program. This will shorten development lead-time for mass-produced products equipped with deep learning capabilities, and enable the mass production of on-device AI solutions. The following hardware platforms are currently supported. KEIm-CVSoC: These SoMs, developed for mass production by Kondo Electronics Industry, are equipped with Intel® FPGAs. They are suited to a wide range of applications from technological studies to product prototyping to mass production. So-One Module: This is the world's smallest module equipped with an MPSoC, developed by PALTEK. The MPSoC, from Xilinx, is well-suited to edge computing.</p>
<h2 id="interview-with-general-manager-yamazaki-on-this-release">Interview with General Manager Yamazaki on This Release</h2>
<p>"We have made full use of our knowledge and experience to develop this commercial version of Efficiera, in order to make people's lives richer and more convenient through deep learning. We are proud that Efficiera has the potential to revolutionize various aspects our daily lives. For example, we will enrich our lives by protecting the safety of small children, supporting secure lifestyles for the elderly, and monitoring normally inaccessible places that form our social infrastructure. In order for Efficiera to be used in such situations, our goal is for it to become an AI standard for on-device solutions smaller even than a smartphone or laptop. To achieve this standard, we have searched for and implemented solutions to a variety of issues encountered in our extensive project experience. One of these is extremely low bit quantization. We established extremely low bit quantization in anticipation of the need for IP cores that require few resources, so that they can be incorporated into small devices. There are few companies with knowledge of both embedded devices (hardware) and trained AI models (software), and that is why this has not yet been put into practical application. I am convinced that LeapMind is a rare company that can bring about practical application at an accelerated pace by approaching the issue from both sides. In the future, we will continue to improve our IP cores to provide suitable performance at ultra-low power. Regarding software, we would like to support general frameworks so that our customers can utilize them along with extremely low bit quantization. Together with FPGA suppliers, OS support, and our partners in AI development, we will further accelerate our activities to secure business and enrich our lives."</p>
<p>1）CNN: Convolutional Neural Network. This type of deep learning is commonly used for image and video recognition.
2）Extremely low bit quantization: Technology developed by LeapMind to reduce the weight of deep learning models. In general, using a numerical representation with a wide bit width such as 32 or 16 bits improves the accuracy of the inference results. However, it also increases the scale (area) of the arithmetic circuit and increases the processing time and power consumption. On the other hand, it is known that narrowing the bit width decreases the size of the circuit and reduces both processing time and power consumption. However, accuracy also decreases, and this has been an issue in achieving both energy efficiency and a small footprint. LeapMind has developed extremely low bit quantization that enables extreme quantization such as 1 bit for weight (weight coefficient) and 2 bits for activation (intermediate data) while maintaining accuracy, resulting in a significant reduction in model size and faster speed, while maximizing power and area efficiency.</p>
<p>*Intel, the Intel logo, the Intel Inside logo, and other names and logos are trademarks of Intel Corporation or its subsidiaries in the United States and/or other countries.
*Product names in this text are trademarks or registered trademarks of LeapMind, Inc.
*The information in this press release is current as of the date of publication.</p>
<h2 id="about-leapmind-inc">About LeapMind, Inc.</h2>
<p>LeapMind, Inc. was founded in 2012 with the corporate philosophy of "bringing new devices that use machine learning to the world". Total investment in LeapMind to date has reached 4.99 billion yen. The company's strength is in extremely low bit quantization for compact deep learning solutions. It has a proven track record of achievement with over 150 companies, centered in manufacturing including the automobile industry. It is also developing its Efficiera semiconductor IP core, based on its experience in the development of both software and hardware.</p>
<p>Head office: Shibuya Dogenzaka Sky Building 5F, 28-1 Maruyama-cho, Shibuya-ku, Tokyo 150-0044
Representative: Soichi Matsuda, CEO
Established: December 2012
URL：<a href="https://leapmind.io/en/">https://leapmind.io/en/</a></p>
<p><strong>Press Relations Contact Information</strong>
LeapMind, Inc.
Takai, Marketing Communication Division
Tel:03-6696-6267
E-mail: <a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>
<p><strong>Inquiries about Products and Services</strong>
LeapMind, Inc.
Business Development Team, Blueoil Division
Tel:03-6696-6267
E-mail:<a href="mailto:business@leapmind.io">business@leapmind.io</a>
Official website:<a href="https://leapmind.io/products/efficiera-ip/">https://leapmind.io/products/efficiera-ip/</a></p>

【Press Release】Official Commercial Launch of Efficiera Ultra-Low Power AI Inference Accelerator IP Core

<h2 id="leapmind-unveils-efficierathe-new-ultra-low-power-ai-inference-accelerator-ip">LeapMind Unveils "Efficiera",
the New Ultra Low Power AI Inference Accelerator IP</h2>
<p>The company drives creation of embedded AI products by providing accelerators featuring superior power and area efficiency</p>
<p>April 20, 2020 – Tokyo, Japan - LeapMind Incorporated (Shibuya-ku, Tokyo Japan, CEO: Soichi Matsuda), the industry leading provider of business solutions built around deep learning technology, today announced that the company has developed "Efficiera", an ultra low power AI inference accelerator IP for targeting companies that design ASIC and FPGA circuits, and other related products. LeapMind plans to ship "Efficiera" in the Fall of 2020. "Efficiera" will enable customers to develop cost-effective, low power edge devices and accelerate go-to-market of custom devices featuring AI capabilities.</p>
<h2 id="overview-of-efficiera-and-related-products">Overview of Efficiera and related products</h2>
<p>"Efficiera" is an Ultra Low Power AI Inference Accelerator IP specialized for Convolutional Neural Network (CNN)(1) inference calculation processing; it functions as a circuit in an FPGA or ASIC device. Its "Extremely low bit Quantization" technology, which minimizes the number of quantized bits to 1–2 bits, does not require cutting-edge semiconductor manufacturing processes or the use of specialized cell libraries to maximize the power and space efficiency associated with convolution operations, which account for a majority of inference processing.</p>
<p>This product enables the inclusion of deep learning capabilities in various edge devices that are technologically limited by power consumption and cost, such as consumer appliances (household electrical goods), industrial machinery (construction equipment), surveillance cameras, and broadcasting equipment as well as miniature machinery and robots with limited heat dissipation capabilities.</p>
<p>LeapMind is simultaneously launching several related products and services: "Efficiera SDK," a software development tool providing a dedicated learning and development environment for Efficiera, the "Efficiera Deep Learning Model" for efficient training of deep learning models, and "Efficiera Professional Services," an application-specific semi-custom model building service based on LeapMind's expertise that enables customers to build extremely low bit quantized deep learning models applicable to their own unique requirements.</p>
<p>Official product website URL：<a href="https://leapmind.io/en/products/efficiera-ip">https://leapmind.io/en/products/efficiera-ip</a></p>
<h3 id="target-applications">Target Applications</h3>
<p><strong>(1) "Hazard Proximity Detection" using object detection</strong>
Helps ensure safety when using industrial vehicles such as construction machinery, by detecting surrounding people and obstacles.</p>
<p><strong>(2)"High quality video streaming" using noise reduction</strong>
Improves image quality by eliminating image noise when shooting under low-light conditions and by blocking noise caused by image codecs.</p>
<p><strong>(3)"Higher resolution for video footage" using super-resolution</strong>
Converts low-resolution video data into resolutions suitable for display devices.</p>
<p><img src="/images/posts/en/news/5086/361bb102-press-efficiera02_en-660x183.png" alt="">
Photo caption: Target applications</p>
<h2 id="advantages-of-efficiera">Advantages of Efficiera</h2>
<p>"Efficiera" offers excellent power and space efficiencies, contributing to power savings and providing cost reductions in AI-equipped products. In addition, because the circuit information is licensed rather than being provided as a module or device, customers can integrate Efficiera within a device featuring other circuits, thereby contributing to reduced BoM(Bill of Material) costs for mass-produced products equipped with AI capabilities. The circuit configuration:</p>
<p><strong>(1)Power Savings</strong>: The power required for convolutional processing is reduced by minimizing the amount of data transfer and the number of bits.</p>
<p><strong>(2)Performance</strong>: The number of calculation cycles can be reduced by reducing the calculation logic, thereby improving calculation performance relative to area and on a per cycle basis.</p>
<p><strong>(3)Space savings</strong>: The silicon area is reduced while maintaining performance by reducing the calculation logic using 1–2 bit quantization; thus, the area per computing unit is minimized.</p>
<p>*The performance estimate (PPA(2)) results during the development stage were measured in cooperation with Alchip Technologies, Limited, and announced at the "COOL Chips 23" international computer conference on April 17th, 2020. These results are featured in the following press release.</p>
<p>▼ LeapMind announces performance results for its in-house developed AI inference accelerator at COOL Chips 23
URL：<a href="https://leapmind.io/en/news/detail/5077/">https://leapmind.io/en/news/detail/5077/</a></p>
<h2 id="development-background">Development background</h2>
<p>Artificial Intelligence technology, particularly deep learning, has received significant attention in recent years. Deep learning has dramatically improved classification accuracy in several fields, including image and audio processing. However, vast computing resources (significant quantities of memory and high processing power) are required to perform the necessary calculations. Hence, it is difficult to perform these calculations on edge devices (terminals) with limitations in power draw, cost, and thermal management.(3) To solve this problem, LeapMind has been engaged in research and development aimed at reducing the load of deep learning models by focusing on a method called "extremely low bit quantization technology," based on which Efficiera was developed. Hereafter, we plan to continue this development with the goal of enabling the incorporation of image recognition capabilities in all types of edge devices.</p>
<h2 id="supporting-efficiera-leapminds-proprietary-extremely-low-bit-quantization-technology">Supporting Efficiera, LeapMind's proprietary "Extremely low bit quantization technology"</h2>
<p>LeapMind is working on the development of a technology to reduce the load of deep learning models, called "extremely low bit quantization technology." In general, the use of numerical expressions with wide bit ranges of 16 or 32 bits (FP16 and FP32 data types) improves the accuracy of inferential results; however, it also increases the size of calculation circuits (area) and both processing time and power consumption. Conversely, when the bit width is reduced and a numerical expression with low bit width of 1–2 bits is used, the circuit scale is reduced and both processing time and power consumption are reduced; however, this results in a decrease in the accuracy, which is an issue when attempting to reduce power and area. However, using extremely low bit quantization technology, LeapMind provides extremely quantization such as a 1-bit Weight (weight coefficient) and 2-bit Activation (Intermediate data), while maintaining accuracy and achieving a significant reduction in the model area, thereby maximizing speed, power efficiency, and space efficiency.</p>
<p><img src="/images/posts/en/news/5086/c1860dba-press-efficiera03_en-660x429.png" alt=""></p>
<h3 id="note-to-editors">Note to editors</h3>
<p><strong>(1)An abbreviation for “Convolutional Neural Network.” This is a type of deep learning architecture that is widely used for image and video recognition.</strong></p>
<p><strong>(2)This is an acronym for “Power, Performance, and Area.”</strong></p>
<p><strong>(3)In particular, power consumption is a considerably important challenge in edge devices with limited power supply.
Moreover, there are several advantages to employing minimal quantization on edge devices, such as reducing delays and ensuring security.</strong></p>
<p>LeapMind and Efficiera are trademarks of LeapMind Incorporated. All other product names noted herein may be trademarks of their respective holders.</p>
<p>This press release may contain forward-looking statements based on Silicon Labs’ current expectations. These forward-looking statements involve risks and uncertainties. A number of important factors could cause actual results to differ materially from those in the forward-looking statements.</p>
<h2 id="about-leapmind-incorporated">About LeapMind Incorporated</h2>
<p>Founded in 2012 with the company philosophy of “promoting new devices using machine learning technologies far and wide,” LeapMind Inc. has procured 4.99 billion yen and specializes in extremely low bit quantization technology that makes deep learning compact. The company’s products are employed in more than 150 businesses primarily in manufacturing, in industries such as the automotive industry. LeapMind Inc. also develops semiconductor IP based on the company’s expertise in software and hardware.</p>
<p>Head Office: 150-0044 3F, Shibuya Dogenzaka Sky Bldg, 28-1 Maruyama-cho, Shibuya-ku, Tokyo, Japan
Representative: Soichi Matsuda, CEO
Established: December 2012
URL:<a href="https://leapmind.io/en/">https://leapmind.io/en/</a></p>
<p><strong>Media Contact</strong>
LeapMind Inc.
Ms. Takai, PR/Branding Division
TEL: 03-6696-6267
EMAIL:<a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>
<p><strong>For inquiries concerning our products and services</strong>
LeapMind Inc.
Blueoil Division　Business Development Team
TEL：03-6696-6267
EMAIL：<a href="mailto:business@leapmind.io">business@leapmind.io</a></p>

LeapMind Unveils “Efficiera”, the New Ultra Low Power AI Inference Accelerator IP

<h2 id="ppa-estimation-performed-in-collaboration-with-a-manufacturer-specializing-in-the-latest-in-ai-chip-layout-design-and-construction">PPA estimation performed in collaboration with a manufacturer specializing in the latest in AI chip layout design and construction</h2>
<p>April 17, 2020 – Tokyo, Japan- LeapMind Incorporated (located in Shibuya-ku, Tokyo Japan, CEO: Soichi Matsuda), an industry leading provider of the business solutions using the core AI technology of deep learning, today announced the results of performance estimations conducted on their AI inference accelerator at the international computer technology symposium, COOL Chips 23, hosted by the IEEE. The silicon mounting using the ASIC design flow was carried out in cooperation with Alchip Technologies, Limited (Taipei, Taiwan, CEO: Johnny Shen).</p>
<h2 id="cool-chips-23-presentation-contents">COOL Chips 23 Presentation Contents</h2>
<p>2012 was a breakthrough year for both deep learning technology and LeapMind, who since this time have been a central driving force of AI related projects at over 150 companies developing and implementing edge devices (terminals) that are not affected by their network environment in terms of power, cost, latency, security, etc</p>
<p>Since deep learning technology generally requires a significant amount of computing resources, its practical applications in edge devices is affected by the following issues regarding power, accuracy, and speed:</p>
<ul>
<li><strong>(1)Power:</strong> Power must be saved owing to the limited supply of power, e.g., battery drive being limited or heat radiation being limited by casing size.</li>
<li><strong>(2)Accuracy</strong>: The technology must be lightweight so that the memory mounted on the edge device can process data while maintaining trained AI model accuracy.</li>
<li><strong>(3)Speed</strong>: Inference processing must be completed within a certain period of time because there is a control target that awaits and uses the inference results.</li>
</ul>
<p>To address these issues, we are working on the development of a core technology called “extremely quantization technology” that operates at both software and hardware levels with a network optimized for practical applications and a dedicated compiler.</p>
<p>The results of these technological developments thus far were presented in the form of performance estimation values. These performance estimations confirmed that our technology can be expected to be effective not only for FPGA devices, the typical target of development, but also for ASIC technology. In further developing this technology, we hope to introduce custom devices with AI functions to the market.</p>
<h2 id="performance-estimation-description-and-results">Performance Estimation Description and Results</h2>
<p>The performance estimation for ASIC technology was conducted with the cooperation of Alchip. When the PPA (Power, Performance, and Area) estimations were conducted (see Note), which are indices for performance when mounted on silicon, Alchip, along with LeapMind, examined and conducted a series of layout trials on multiple memory configurations. The results of the conducted power simulations, which followed the set layout and wiring, confirmed that the PPA predicted by LeapMind could be achieved.</p>
<p><img src="/images/posts/en/news/5077/d4d02bb5-_en-660x440.png" alt="AI Inference Accelerator IP, PPA Estimated results. Cell library used: TSMCC 12nm 6T SVt. Power consumption (Typical conditions): 27.7 TOP/W. Performance: 6.55 TOP/s @ 800MHz. silicon area: 0.442 mm²."></p>
<h2 id="applications">Applications</h2>
<p>Based on the results of these performance estimations, LeapMind will continue its development of technology for ultra-low power AI inference accelerator circuits, including work on improvements in performance and power efficiency. Furthermore, considering that these performance estimations did not use any special cell library or memory, nor any power-saving methods of implementation, there is still room for improving power efficiency at the implementation stage.</p>
<p>For example, by adopting design techniques from our collaborating partner, Alchip, a company that specializes in state-of-the-art layout designing and manufacturing of AI/HPC chips, not only both static and dynamic power consumptions can be predicted and reduced, but also further reduction of power by using clock gating techniques and an effective division of area to account for multiple power supply voltages are possible.</p>
<p>Moreover, since the accelerator can be adapted for feature re-extraction for use under ultra-low voltage conditions, we believe that even greater performance improvements can be expected in actual use with ASIC technology.</p>
<h2 id="about-leapmind-incorporated">About LeapMind Incorporated</h2>
<p>Founded in 2012 with the company philosophy of “promoting new devices using machine learning technologies far
and wide,” LeapMind Inc. has procured 4.99 billion yen and specializes in extremely low bit quantization technology that makes deep learning compact. The company’s products are employed in more than 150 businesses primarily in manufacturing, in industries such as the automotive industry. LeapMind Inc. also develops semiconductor IP based on the company’s expertise in software and hardware.</p>
<p>Head Office: 150-0044 3F, Shibuya Dogenzaka Sky Bldg, 28-1 Maruyama-cho, Shibuya-ku, Tokyo, Japan
Representative: Soichi Matsuda, CEO
Established: December 2012
URL:<a href="https://leapmind.io/en/">https://leapmind.io/en/</a></p>
<h2 id="about-alchip-technologies-limited">About Alchip Technologies, Limited</h2>
<p>Alchip Technologies Ltd, headquartered in Taipei, Taiwan, is a global leading provider of silicon design and production services for system companies developing complex and high-volume ASICs and SoCs. Alchip was founded by semiconductor veterans from Silicon Valley and Japan in 2003 and provides faster time-to-market and cost-effective solutions for SoC design at mainstream and advanced, including 7nm processes. Customers include global leaders in AI, HPC/supercomputer, mobile phones, entertainment device, networking equipment and other electronic products. Alchip is listed on the Taiwan Stock Exchange (TWSE: 3661).</p>
<p>Head Office: 9F., No.12, Wenhu St., Neihu Dist., Taipei, Taiwan 114
Representative: Johnny Shen, CEO
Established: February 2003
URL:<a href="http://www.alchip.com">http://www.alchip.com</a></p>
<p><strong>Media Contact</strong>
LeapMind Inc.
Ms. Takai, PR/Branding Division
TEL: 03-6696-6267
EMAIL:<a href="mailto:pr@leapmind.io">pr@leapmind.io</a></p>

News