Unlike server-based systems, on-device environments operate under strict memory and computing constraints. As a result, reducing AI model size and maximizing runtime efficiency are essential. To meet this challenge, Samsung Research AI Center is leading work across core technologies — from model compression and runtime software optimization to new architecture development.

Samsung Newsroom sat down with Dr. MyungJoo Ham, Master at AI Center, Samsung Research, to discuss the future of on-device AI and the optimization technologies that make it possible.

The First Step Toward On-Device AI

At the heart of generative AI — which interprets user language and produces natural responses — are large language models (LLMs). The first step in enabling on-device AI is compressing and optimizing these massive models so they run smoothly on devices such as smartphones.

“Running a highly advanced model that performs billions of computations directly on a smartphone or laptop would quickly drain the battery, increase heat and slow response times — noticeably degrading the user experience,” said Dr. Ham. “Model compression technology emerged to address these issues.”

LLMs perform calculations using extremely complex numerical representations. Model compression simplifies these values into more efficient integer formats through a process called quantization. “It’s like compressing a high-resolution photo so the file size shrinks but the visual quality remains nearly the same,” he explained. “For instance, converting 32-bit floating-point calculations to 8-bit or even 4-bit integers significantly reduces memory use and computational load, speeding up response times.”

A drop in numerical precision during quantization can reduce a model’s overall accuracy. To balance speed and model quality, Samsung Research is developing algorithms and tools that closely measure and calibrate performance after compression.

“The goal of model compression isn’t just to make the model smaller — it’s to keep it fast and accurate,” Dr. Ham said. “Using optimization algorithms, we analyze the model’s loss function during compression and retrain it until its outputs stay close to the original, smoothing out areas with large errors. Because each model weight has a different level of importance, we preserve critical weights with higher precision while compressing less important ones more aggressively. This approach maximizes efficiency without compromising accuracy.”

Beyond developing model compression technology at the prototype stage, Samsung Research adapts and commercializes it for real-world products such as smartphones and home appliances. “Because every device model has its own memory architecture and computing profile, a general approach can’t deliver cloud-level AI performance,” he said. “Through product-driven research, we’re designing our own compression algorithms to enhance AI experiences users can feel directly in their hands.”

The Hidden Engine That Drives AI Performance

Even with a well-compressed model, the user experience ultimately depends on how it runs on the device. Samsung Research is developing an AI runtime engine that optimizes how a device’s memory and computing resources are used during execution.

“The AI runtime is essentially the model’s engine control unit,” Dr. Ham said. “When a model runs across multiple processors — such as the central processing unit (CPU), graphics processing unit (GPU) and neural processing unit (NPU) — the runtime automatically assigns each operation to the optimal chip and minimizes memory access to boost overall AI performance.”

The AI runtime also enables larger and more sophisticated models to run at the same speed on the same device. This not only reduces response latency but also improves overall AI quality — delivering more accurate results, smoother conversations and more refined image processing.

“The biggest bottlenecks in on-device AI are memory bandwidth and storage access speed,” he said. “We’re developing optimization techniques that intelligently balance memory and computation.” For example, loading only the data needed at a given moment, rather than keeping everything in memory, improves efficiency. “Samsung Research now has the capability to run a 30-billion-parameter generative model — typically more than 16 GB in size — on less than 3 GB of memory,” he added.

The Next Generation of AI Model Architectures

Research on AI model architectures — the fundamental blueprints of AI systems — is also well underway.

“Because on-device environments have limited memory and computing resources, we need to redesign model structures so they run efficiently on the hardware,” said Dr. Ham. “Our architecture research focuses on creating models that maximize hardware efficiency.” In short, the goal is to build device-friendly architectures from the ground up to ensure the model and the device’s hardware work in harmony from the start.

Training LLMs requires significant time and cost, and a poorly designed model structure can drive those costs even higher. To minimize inefficiencies, Samsung Research evaluates hardware performance in advance and designs optimized architectures before training begins. “In the era of on-device AI, the key competitive edge is how much efficiency you can extract from the same hardware resources,” he said. “Our goal is to achieve the highest level of intelligence within the smallest possible chip — that’s the technical direction we’re pursuing.”

Today, most LLMs rely on the transformer architecture. Transformers analyze an entire sentence at once to determine relationships between words, a method that excels at understanding context but has a key limitation — computational demands rise sharply as sentences get longer. “We’re exploring a wide range of approaches to overcome these constraints, evaluating each one based on how efficiently it can operate in real device environments,” Dr. Ham explained. “We’re focused not just on improving existing methods but on developing the next generation of architectures built on entirely new methodologies.”

The Road Ahead for On-Device AI

What is the most critical challenge for the future of on-device AI? “Achieving cloud-level performance directly on the device,” Dr. Ham said. To make this possible, model optimization and hardware efficiency work closely together to deliver fast, accurate AI — even without a network connection. “Improving speed, accuracy and power efficiency at the same time will become even more important,” he added.

Advancements in on-device AI are enabling users to enjoy fast, secure and highly personalized AI experiences — anytime, anywhere. “AI will become better at learning in real time on the device and adapting to each user’s environment,” said Dr. Ham. “The future lies in delivering natural, individualized services while safeguarding data privacy.”

Samsung is pushing the boundaries to deliver more advanced experiences powered by optimized on-device AI. Through these efforts, the company aims to provide even more remarkable and seamless user experiences.

Central to Samsung’s AI innovations is on-device AI. This technology enables AI to operate independently within the device and does not require a server or the cloud. Users benefit from quick AI response times without needing a network connection, reducing concerns over personal data leakage.

On-device AI has now evolved beyond simple functions to handle demands requiring vast amounts of data — such as generative AI — using only the device’s internal resources.

Samsung Newsroom explores how Samsung is leading the world in on-device AI innovations.

No Cloud Required: Transforming the On-Device AI Experience

Citing the need for speed and security, much of the tech industry has focused on running complex generative AI in an on-device environment. With long-term investments that date back more than a decade, Samsung has already customized AI features for many of its devices and positioned itself front and center in providing on-device AI experiences for everyday life.

Samsung’s Galaxy AI, for example, offers on-device capabilities across the company’s latest mobile devices and tablets.¹ Now supporting a total of 20 languages,² Galaxy AI enables communication without language barriers including real-time conversation interpretation as well as message and web page translation.

For TVs, Samsung leverages its expertise as a global leader in the market to train AI models. Since 2020, the company has continuously enhanced its TV-specific processor with a built-in NPU (neural processing unit). The 2024 Neo QLED 8K model’s NQ8 AI Gen3 processor features 512 neural networks for enhanced image and sound quality. By analyzing and modifying pixels, frames and sound sources, the processor delivers superior upscaling, smooth motion and clear dialogue for a seamless viewing experience in all situations.

The key to on-device AI is minimizing model size while maintaining performance. To achieve this, Samsung has focused on efficient implementation while utilizing specialized AI training data tailored to specific tasks.

▲ Sunmin Kim and Cheulhee Hahm, Master, Visual Display Business, Samsung Electronics explain on-device AI.

Smaller and Faster: Samsung’s AI Optimization Research

As AI technology advances and data processing requirements increase, model optimization has become crucial. In other words, high-performance AI models must be able to operate efficiently within the limits of the devices’ resources — including their processors, memories and batteries — while retaining performance levels and reliability for daily use. Key technologies for this include model compression, hardware optimization and data processing acceleration.

Samsung has secured various technologies in the field of model compression to reduce the size of AI models. The company has succeeded in developing light and fast AI models through techniques like quantization, which enhances response speed by simplifying algorithms and optimizing computation processes; pruning, which removes non-essential elements of a larger model; and knowledge distillation, which transfers knowledge from large models to smaller models.

Samsung also continues to innovate in hardware optimization and data processing acceleration to swiftly and efficiently run AI. For example, flash utilization technology partitions large AI models and significantly reduces memory usage. Additionally, the company has developed patented technology that is expected to assist quick inference on low-end devices without an NPU to broaden AI applications across various products. Further research is underway on increasing inference speeds with speculative decoding, a technology that predicts AI model outcomes, and enhancing computing power by running various hardware simultaneously.

▲ Pruning and knowledge distillation

End-to-End Expertise: From In-house Development to Implementation

With expertise spanning hardware, software, components and end products, Samsung is spearheading on-device AI innovation from research and development to product implementation.

From a device perspective, Samsung’s market leadership in mobile devices, TVs, home appliances and more represents its excellence in both hardware and software. Samsung integrates AI into its own devices, allowing the company to leverage both in-house development and open collaboration with industry partners to create hardware-optimized AI models.

Samsung’s expertise also includes system software to bridge hardware and application software. For example, the Tizen operating system and the NPUs on Samsung TVs power on-device AI for a smarter, more enhanced viewing experience. The company also provides developers with Vision AI and Language AI SDKs, as well as machine learning APIs to facilitate AI model training and inference.

Samsung Research, the company’s advanced research and development hub, is focused on enhancing competitiveness by utilizing specialized expertise from its global network of labs — from AI model structure development and model compression in the U.K. and U.S. to creating AI acceleration solutions in India and Poland.

From a components perspective, Samsung has endlessly driven progress in NPU performance as a leading manufacturer of semiconductors and plans to continue working with its partners to develop new ways of utilizing on-device AI.

▲ Samsung introduced a new AI TV experience at Samsung Developer Conference 2024 in October.

With its on-device AI capabilities expected to expand across more products and services, Samsung is pioneering groundbreaking changes that will see AI technologies transform daily life. In part two of this series, Samsung Newsroom will explore the company’s innovations in personalized AI.

¹ Galaxy AI-supported devices include the Galaxy Z Fold6, Flip6, Fold5, Flip5, Fold4, Flip4, S24 series, S23 series, S22 series, S24 FE, S23 FE, Tab S10 series, Tab S9 series, Tab S8 series, etc.
² Galaxy AI-supported languages include Korean, English, Arabic, Simplified Chinese, Dutch, French, German, Hindi, Indonesian, Italian, Japanese, Polish, Portuguese, Romanian, Russian, Spanish, Swedish, Thai, Turkish and Vietnamese.

Samsung Electronics, the world leader in advanced memory technology, today announced it has successfully developed the industry’s first PCIe 4.0 automotive SSD based on eighth-generation vertical NAND (V-NAND). With industry-leading speeds and enhanced reliability, the new auto SSD, AM9C1 is an optimal solution for on-device AI capabilities in automotive applications.

With about 50% improved power efficiency compared to its predecessor, the AM991, the new 256GB auto SSD will deliver sequential read and write speeds of up to 4,400 megabytes-per-second (MB/s) and 400MB/s, respectively.

“We are collaborating with global autonomous vehicle makers and providing high-performance, high-capacity automotive products,” said Hyunduk Cho, Vice President and Head of Automotive Group at Samsung Electronics’ Memory Business. “Samsung will continue to lead the Physical AI¹ memory market that encompasses applications from autonomous driving to robotics technologies.”

Built on Samsung’s 5-nanometer (nm) controller and providing a single-level cell (SLC) Namespace² feature, the auto SSD AM9C1 demonstrates high performance for easier access to large files. By switching from its original triple-level cell (TLC) state to SLC mode, users can enjoy boosted read and write speeds of up to 4,700MB/s and 1,400MB/s, respectively, while also benefiting from the added reliability of SLC SSDs.

The 256GB AM9C1 is currently being sampled by key partners and is expected to begin mass production by the end of this year. Samsung plans to offer multiple storage capacities for the AM9C1 ranging from 128GB to 2 terabytes (TB) to address the growing demand for high-capacity automotive SSDs. The 2TB model, which is set to offer the industry’s largest capacity in this product category, is scheduled to start mass production early next year.

Through intensified board-level tests, Samsung’s new auto SSD satisfies the automotive semiconductor quality standard AEC-Q100³ Grade 2, ensuring stable performance over a wide temperature range of -40°C to 105°C.

To further meet the high standards of the automotive industry in terms of durability and stability, Samsung also conducts various quality assurance processes. The company received ASPICE CL3 authentication⁴ for its UFS 3.1 product in March this year.

In an effort to obtain CSMS certification based on ISO/SAE 21434,⁵ Samsung will continue to actively enhance the technological reliability and stability of its automotive solutions.

“ASPICE and ISO/SAE 21434 certifications are milestones that affirm the reliability and stability of our technology,” said Hwaseok Oh, Executive Vice President at Samsung Electronics’ Memory Business. “Beyond these achievements, Samsung will continue to elevate its product stability and quality by consistently providing the best solution to key partners.”

¹ AI such as robots and autonomous vehicles that perceives and interacts with the physical world through sensors.
² Provides SLC partition with better performance and reliability than TLC, allowing users to configure it in accordance to data type. However, when switched to SLC mode capacity decreases to 1/3 of the TLC.
³ Global standard that Automotive Component Manufacturers Association (ACMA) has established for the reliability evaluation procedures and criteria for automotive electronic components.
⁴ Automotive Software Process Improvement and Capability dEtermination (ASPICE) is a software development standard developed and distributed by the German Automotive Association (VDA) that evaluates the reliability and competence of automotive component manufacturers’ software development processes. It is divided into Capability Level (CL) stages 0 to 5, with CL3 meaning that an organization has established a systematic process and can effectively execute it.
⁵ Cyber Security Management System certification is an international standard designed to enhance cybersecurity in the automotive industry based on ISO/SAE 21434, covering cybersecurity processes and requirements from design to development, evaluation and mass production.

Samsung Electronics has transformed people’s lives by popularizing wall-mounted TVs and smart TVs. Long at the forefront of innovation in TV hardware and software, Samsung is now utilizing AI to overcome limitations in the viewing experience and define the next generation of screen experiences.

▲ SW Yong, President and Head of Visual Display Business, Samsung Electronics

How AI Screens are Changing Daily Life

In today’s hyperconnected era, the AI TV is set to become the hub of an AI-powered smart home system — connecting, managing and controlling devices to provide consumers with new levels of convenience for more valuable experiences. Serving as a command center for the AI home, the AI TV allows users to easily monitor and operate connected devices, which can be automatically integrated to the SmartThings ecosystem via Calm Onboarding technology.

To see appliances at a glance, users can simply check the Daily Board on their Samsung AI TVs. The large screen displays 3D Map View to show the status of devices scattered around the home. Meanwhile, energy consumption can be managed by remotely turning off the lights or air conditioner when left on. Cameras and sensors ensure the home is monitored from afar, and users can even play videos to keep their pets entertained.

▲The 8K NQ8 AI Gen3 Processor combines AI sound, AI picture and AI optimization features for a frictionless viewing experience

In addition, AI TVs are offering an ultra-large, high-resolution viewing experience to make users feel as if they are in first-row VIP seats at a stadium, movie theater or concert hall.

The Neo QLED 8K offers 8K AI Upscaling Pro so users can enjoy SD/HD/Full HD high-resolution images at 8K levels. AI Motion Enhancer Pro makes sports content feel even more real and immersive. Meanwhile, Active Voice Amplifier Pro helps users catch every word — from dialogue in a movie amid background music to a sportscaster’s commentary over cheers from a crowd. With these features, the world is entering the era of the AI home cinema and the AI home sports arena.

Furthermore, AI TVs will help make TV more accessible to all. Optical character recognition technology converts on-screen text to audio subtitles in real time for users who are hard of hearing. On the other hand, Relumino Mode configures contents on screen by amplifying contrast, brightness and contours for users with low vision.

AI Screens Lead the Evolution of AI Homes

AI TVs are making new experiences possible — experiences that would have been unimaginable with conventional TVs. To this end, AI homes will evolve with Samsung’s Tizen operating system.

With AI becoming a differentiator in the global home appliance market, operating systems are a key player when incorporating AI into TVs.

Since 2015, Tizen OS has powered hundreds of millions of smart TVs and led the evolution of TVs. Thanks to the platform’s excellent performance, Samsung expanded its capabilities from TV hardware to software.

▲Since its launch in 2015, Tizen has become a dominant OS for smart TVs

Equipped with SmartThings and Tizen OS, Samsung AI TVs will increasingly tailor their services to users’ tastes and lifestyles. Boundaries separating products from service platforms will crumble away to open the ‘Era of Device Platform Innovation.’ Continued upgrades to Tizen OS will ensure quality service in the future.

The key to these personalized, customized experiences offered by AI TVs is on-device AI — in other words, AI that is embedded in the TV with no need for a cloud connection. This technology is more secure than cloud-based AI because personal information wouldn’t be sent out to a server.

Above all, ‘Samsung Knox,’ Samsung’s proprietary security solution, will thoroughly protect devices and personal information so users can enjoy a safe AI home.

The Future of AI: What This Means for Consumers

The rapid evolution of AI raises many social, ethical, economic and legal concerns. But ultimately, AI will evolve in a way that provides greater value for consumers.

Samsung strives to create a new way of life with AI screens by understanding, responding and adjusting to users’ needs — for new screen experiences that were not possible before.

Whether breaking down language barriers during communication, sharing content recommendations or controlling home appliances, AI is already impacting all of our lives. Through continued improvements to its AI screens, Samsung will create a more valuable daily life in the AI home.

Samsung Electronics, the world leader in advanced memory technology, today announced it has begun mass production for the industry’s thinnest 12 nanometer (nm)-class, 12-gigabyte (GB) and 16GB LPDDR5X DRAM packages, solidifying its leadership in the low-power DRAM market.

Leveraging its extensive expertise in chip packaging, Samsung is able to deliver ultra-slim LPDDR5X DRAM packages that can create additional space within mobile devices, facilitating better airflow. This supports easier thermal control, a factor that is becoming increasingly critical especially for high-performance applications with advanced features such as on-device AI.

“Samsung’s LPDDR5X DRAM sets a new standard for high-performance on-device AI solutions, offering not only superior LPDDR performance but also advanced thermal management in an ultra-compact package,” said YongCheol Bae, Executive Vice President of Memory Product Planning at Samsung Electronics. “We are committed to continuous innovation through close collaboration with our customers, delivering solutions that meet the future needs of the low-power DRAM market.”

With the new LPDDR5X DRAM packages, Samsung offers the industry’s thinnest 12 nm-class LPDDR DRAM in a 4-stack structure,¹ reducing the thickness by approximately 9% and improving heat resistance by about 21.2%, compared to the previous generation product.

By optimizing printed circuit board (PCB) and epoxy molding compound (EMC)² techniques, the new LPDDR DRAM package is as thin as a fingernail at 0.65 millimeters (mm), the thinnest among existing LPDDR DRAM of 12GB or above. Samsung’s optimized back-lapping³ process is also used to minimize the package height.

Samsung plans to continue expanding the low-power DRAM market by supplying its 0.65mm LPDDR5X DRAM to mobile processor makers as well as mobile device manufacturers. As demand for high-performance, high-density mobile memory solutions in smaller package sizes continues to grow, the company plans to develop 6-layer 24GB and 8-layer 32GB modules into the thinnest LPDDR DRAM packages for future devices.

¹ Structure with four layers packaged together, and each layer consists of two LPDDR DRAMs.
² Material that protects semiconductor circuits from various external environments such as heat, impacts, and moisture.
³ Grinding the backside of a wafer.

Samsung Electronics today announced that it has signed an agreement to acquire Oxford Semantic Technologies, a UK startup specializing in knowledge graph technology.

Established in 2017 by three Oxford University professors — Ian Horrocks, Boris Motik and Bernardo Cuenca Grau — Oxford Semantic Technologies holds cutting-edge technological capabilities in the areas of knowledge representation and semantic reasoning. With its AI-centric engine, RDFox®, the company currently collaborates with organizations across Europe and North America involved in the finance, manufacturing and e-commerce sectors.

Knowledge graph technology stores information as an interconnected web of related ideas and process data in a manner similar to how humans acquire, remember, recall and reason over knowledge. By integrating and connecting data, this technology enhances the understanding of how people use a product or service and enables rapid information retrieval and recommendation. As such, it is considered one of the key technologies for realizing more sophisticated and personalized AI solutions.

Knowledge graph technology is challenging to implement due to the complex computations used in the process of converting dynamic and extensive real-world data into knowledge graphs and utilizing them. However, Oxford Semantic Technologies has developed and successfully commercialized knowledge graph technology that optimizes data processing and enables advanced reasoning, in the cloud and on-device.

Samsung Electronics has been collaborating with Oxford Semantic Technologies on various projects since 2018, including investment of Samsung Ventures. With this acquisition, Samsung will secure advanced core engines for personal knowledge graphs. These graphs integrate information and context dispersed across various services and apps, crafting a user experience that is tailor-made by becoming increasingly familiar with users’ preferences and usage.

Combined with on-device AI technology, such as that on Samsung’s Galaxy S24 series, personal knowledge graph technology facilitates hyper-personalized user experiences while ensuring sensitive personal data remains secure on the device. It will be applicable across all of Samsung’s products, extending beyond just mobile devices to televisions and home appliances.

“We are delighted to be working with Samsung,” said Peter Crocker, CEO of Oxford Semantic Technologies. “By integrating Samsung’s expertise in user experience and data with our advanced knowledge graph and reasoning technology, we will provide Samsung’s customers with even more sophisticated personalization. In addition, developing RDFox with Samsung, and being part of the larger group, will provide all of our clients with an even better product, service and support.”

“As global consumers realize their growing need for more personalized AI experiences, the acquisition of Oxford Semantic Technologies will further boost Samsung’s strong capabilities in knowledge engineering,” said Paul Kyungwhoon Cheun, Head of Samsung Research and CTO of Samsung Electronics. “This acquisition is another important step forward as we strive to deliver tailor-made AI experiences built upon our hallmark technological innovation.”

About Oxford Semantic Technologies Ltd.
Oxford Semantic Technologies is a spin out of the University of Oxford by leading academics in the field of knowledge representation and reasoning. With its innovative software RDFox®, the company helps leading companies across many sectors to integrate diverse data sets and apply reasoning across them. Use cases include recommendation engines, product configurators, integration of maps for autonomous vehicles and anomaly detection in financial transactions. For more information, please visit www.oxfordsemantic.tech.

We are at the beginning of the most exciting and probably one of the most historic moments in technology. The era of mobile AI is here — which for Samsung opened up by the introduction of Galaxy AI features on the Galaxy S24 series, our first AI phone.

Revolutionizing Mobile Experiences With a Hybrid AI Approach

To bring the real-life benefits of generative AI technology to the Galaxy S24 series, we took a hybrid approach to our AI integration. AI offers so many possibilities, and we believe mobile devices are the primary access point for unlocking its potential as users around the globe continue to rely on their phones for most of their day-to-day needs.

The essential role that mobile devices play in everyday life means they are very personal items that people use during many critical and memorable moments. We understand these behaviors and believe smartphones should work harder so that users can truly make the most of important moments in a simpler, more intuitive and more epic way. We also recognize the importance of privacy which is why we give users full control over what they share and what they keep private.

We believe our hybrid approach is the most practical and reliable solution to meet all these needs and puts Samsung ahead of the curve. We are providing users with a balance between the instant responsiveness and extra privacy assurance of on-device AI and the versatility of cloud-based AI through open collaborations with industry-leading partners in offering a variety of functions they need for daily life.

Expanding Possibilities With On-Device AI

We made the Live Translate feature on-device AI based because voice calls are the most basic feature of a smartphone, yet also an intimate and private means of communication. Samsung has committed to empowering users to communicate without language barriers, while at the same time ensuring every communication is safe and private.

To make this feature a reality, our R&D team worked day and night as part of a collaborative process that truly brought out the best in our teams. From determining the right size of AI language models to training and testing in real-life scenarios, our R&D organizations within the MX business challenged themselves and pushed beyond the limits of what we thought was possible to get this feature running entirely on-device.

Samsung’s global R&D networks around the world, which are in charge of growing the best talents and developing core technologies in each region, have also played a pivotal role. Samsung Research Centers around the globe, including Poland, China, India and Vietnam, have dedicated themselves to developing and expanding the languages supported by Galaxy AI. Languages are very cultural, contemporary and localized, so to help people break down language barriers and communicate more naturally, the role of our local R&D offices is even more crucial in opening up new possibilities for users in more regions than ever before.

All these efforts have enabled Samsung to offer something completely new which I’m excited to share. Soon, Samsung is extending the power of Galaxy AI beyond Samsung’s own native calling app by expanding Live Translate to other third-party message apps to support voice calls. So you can stay in contact with friends or colleagues, communicating on your favorite apps in multiple languages.

And since this feature has been integrated into our on-device AI language translation model, users will be able to experience barrier-free communication without worrying about privacy issues like personal data being shared outside their phone when using Live Translate.

I expect that more and more mobile AI features will be incorporated inside mobile devices as the intelligent computing power of the latest chips — particularly NPUs — advances at a blazing fast speed, along with other technologies. This will enable more people to embrace AI and make every day more convenient while offering greater peace of mind.

Growing Galaxy AI for Tomorrow’s Needs, Today

This exciting news is all part of the extension of our commitment to Galaxy AI and entering the next phase of the mobile AI era. The introduction of Galaxy AI on the Galaxy S24 series was only the start. To deliver a completely new and unique AI experience, we will further optimize the Galaxy AI experience for the upcoming foldable devices. Our foldables are the most versatile and flexible form factor in Samsung Galaxy and when combined with Galaxy AI, these two complementary technologies will together unlock all new possibilities.

Samsung is also committed to expanding Galaxy AI experiences across the broader Galaxy ecosystem in a way that only Samsung can. As the mobile AI era moves forward at a rapid pace, Samsung is accelerating mobile AI innovations to meet the needs of not only today, but tomorrow.

Samsung Research combined data and cutting-edge technology. Together with the Mobile eXperience (MX) Business R&D Office, they further honed this technology to develop the translation features powered by on-device AI — which can be used for real-time translation during calls and across various applications. Samsung Newsroom met with Yoonjung Choi and Yonghyun Ryu from Samsung Research’s Global AI Center to learn more about these ambitious features.

▲ (From left) Yoonjung Choi and Yonghyun Ryu of Samsung Research’s Global AI Center

Smooth Communication and Strong Security With On-Device AI

On-device AI is the key differentiator to Galaxy AI’s Interpreter and Live Translate features — introduced to users through the release of the Galaxy S24 series earlier this year. Leveraging the advanced computing resources built into the devices themselves, smartphones with on-device AI can provide services without relying on servers or the cloud. Users can be assured that their data will remain private and secure since information is not shared with external sources.

Samsung Research’s Global AI Center contributed to integrating proprietary technology into its AI translation model to commercialize these features for widespread use.

Envisioning a wide range of applications, the Samsung Research team and MX R&D Office obtained an expansive amount of data. “We collected colloquial data for real-time translation during calls in Live Translate and travel-related data for Interpreter,” explained Yoonjung Choi, who led the project. “To provide the most accurate translation, we studied and incorporated casual language used in chatrooms and HTML tags used in web browsers.”

▲ Use cases for on-device AI translation include ① Live Translate, ② text message and conversation translation, ③ translation in Samsung Notes and ④ web browser translation.

How Samsung Research Trained Its AI Model

The Samsung Research team’s AI translation model is based on deep learning technology that learns from its own data. Yonghyun Ryu, who is in charge of AI research and development, likened this process to raising a child. “Similar to how a child needs excellent educational resources and caregivers to grow and thrive, good language data and talented researchers are required when developing a high-performance AI translation model,” he described.

Samsung Research has both — since 2013, the company has been providing in-house translation services, conducting R&D related to AI translation and accumulating high-quality data.

Samsung Research’s team of deep learning experts played an important role in training the AI model. “If incorrect translations occur during the research and development process, it is necessary to identify the problem and make improvements. However, this can be challenging and time-consuming for researchers without sufficient capabilities and experience,” he explained. “Our researchers used their expertise and know-how to quickly analyze the cause of the issue and come up with a solution to enhance the AI translation model.”

▲ Yonghyun Ryu from Samsung Research’s Global AI Center

To assess the performance of Galaxy AI’s translations, the Samsung Research team used quantitative metrics based on test sets as well as qualitative evaluations by human translators and the MX R&D Office.

In addition, the team gained credibility by competing in global machine translation competitions. “Although participation requires time and effort, good performance in competitions provides momentum for research and development,” Ryu emphasized. “We were able to achieve strong results because our team members could freely discuss new ideas and put them to the test.”

Politeness and Punctuation: Understanding the Quirks of Each Language

Each language has characteristics that are unique to its culture — these can include honorifics, tonal inflections and distinct punctuation symbols. To make translations as accurate as possible, the AI translation model considers all these linguistic idiosyncrasies. For example, in Korean and Japanese, honorifics are translated to maintain a respectful tone.

Samsung collaborated with regional R&D centers to fully understand languages in their cultural context. “By working closely with researchers and linguists in other countries, we were able to offer a more accurate and complete translation,” said Choi.

At the same time, dealing with different languages often involves trial and error. “Vietnamese, for instance, is a tonal language. However, we realized during the research process that Vietnamese users often omit tones in casual conversations when chatting,” she explained. “We needed additional data to help the features translate sentences without tones.” For Thai, Samsung Research developed a special sentence separator because the language does not use punctuation.

Why Lightweight Technology Is the Key to Effective On-Device AI Models

Samsung Research began developing its on-device AI translation model in 2019. “Unlike server-based AI models, on-device AI models must be driven only using users’ devices,” stated Ryu. “Developing lightweight technology that uses minimal resources is key.” To make the model lighter, the team used “knowledge distillation” and “quantization” technology.

Knowledge distillation is a method that extracts knowledge from a large, high-performing teacher model and delivers it to a smaller student model. This is similar to how a teacher summarizes a topic so that a student can digest the information more easily and efficiently.

Quantization simplifies AI algorithms to reduce model size and streamlines the process to increase response speed.

Ryu compared quantization to drawing strawberries. “You need a wide range of colors to paint lifelike strawberries — but strawberries can also be made with just red and green,” he explained. “Quantization is the process of minimizing the number of colors needed to draw strawberries while trying to make it as close as possible to the real thing.”

▲ Yoonjung Choi from Samsung Research’s Global AI Center

In the AI field, knowledge distillation and quantization are well-known approaches to making models lighter. Yet, implementing them on a commercial scale is not easy due to differences in each researcher’s detailed experimental methods and factors. Samsung Research developed proprietary technology by discovering an efficient quantization technique and creating a accelerated algorithm based on it. “Through constant experimentation, we found an optimal way to make the model lighter,” commented Choi.

By combining the high-quality AI translation model with algorithms that make models lighter and speed up response time, lightweight and fast on-device AI features were born.

The Culture Behind the Language: What Would the Perfect AI Translation Model Look Like?

The researchers at Samsung Research’s Global AI Center have bright goals as they lead the field of on-device AI. “My ultimate goal is to help users communicate smoothly and conveniently with people who speak other languages,” said Choi. Ryu revealed his vision to build the perfect translator. “One day, we want to create a translator that truly understands the cultures behind the languages it is translating, equipped with an extensive pool of knowledge,” he expressed. “I want to challenge myself to develop a translator the world has never seen before.”

▲ Samsung Research’s Global AI Center team developed the on-device AI translation feature.

Samsung’s on-device AI translation features allow anyone with an enabled mobile device to communicate freely — without worrying about internet connection or information leakage. Going forward, the Samsung Research team will continue to spearhead innovations in the rapidly evolving field of AI and bring new levels of convenience to users.

Speakers representing various fields introduced newly developed AI algorithms, as well as innovative AI solutions that can benefit our lives in the future. Samsung Electronics livestreamed this year’s forum on its YouTube channel and offered participants – which included engineers, researchers and students in the field of AI – the opportunity to interact with experts during a Q&A session.

Read on for Samsung Newsroom’s recap of the presentations and key topics that took center stage during the two-day event.

Samsung AI Forum Day One

Developing AI That Addresses Common Problems

Hosted by the Samsung Advanced Institute of Technology (SAIT), Samsung’s R&D hub dedicated to cutting-edge future technologies, day one of the Samsung AI Forum began with opening remarks from Dr. Kinam Kim, Vice Chairman and CEO of Samsung Electronics. “Digital transformation has been accelerated in every industry, to which data science and machine learning are essential,” said Dr. Kim. “We at Samsung are open to discussing how to tackle important, common problems with researchers from all over the world, and we hope that the Samsung AI Forum can help facilitate that goal.”

This was followed by a keynote speech from Professor Yoshua Bengio of the University of Montreal, the co-chair of the Samsung AI Forum and a Samsung AI Professor. During his speech, Professor Bengio introduced a new machine learning tool called GFlowNets.

After explaining how the algorithms could be applied to the development of new drugs, he emphasized how “We find that [the model] converges to good solutions faster than other methods, and in addition, it finds a more diverse set of solutions. So this is very encouraging, and we are very excited about the potential applications in discovery in general.” After finishing his speech, the professor discussed ways to apply the algorithms during a Q&A session that featured scientists from around the world.

The keynote was followed by three technology sessions entitled “Scalable & Sustainable AI Computing”, “AI for Scientific Discovery” and “Trustworthy Computer Vision”. During these sessions, leading academics and startups spoke alongside some of Samsung’s top researchers.

Professors Kunle Olukotun of Stanford University, Gerbrand Ceder of the University of California – Berkeley and Antonio Torralba of the Massachusetts Institute of Technology shared key findings in their respective areas of AI research. Founders of startups based in Silicon Valley, including Andrew Feldman, CEO of Cerebras Systems, Bryce Meredig, CSO of Citrine Informatics and Daniel Bibireata, Vice President of Landing AI, presented insights on business models for various areas of AI research, as well as future business strategies. Representing Samsung were multiple leading researchers, including Changkyu Choi, Senior Vice President and Head of SAIT’s AI & SW Research Center, who introduced the company’s vision for AI and summarized the progress it has made through its research in the field.

The event also showcased rising talents and researchers in the field of AI. Samsung revealed this year’s five winners of the Samsung AI Researcher of the Year award, which was launched last year to recognize promising global AI researchers.

“I’m especially thankful to my students, whose work is really what’s being rewarded here,” said Professor Phillip Isola of the Massachusetts Institute of Technology, who received the award. “We’re trying to make progress to make AI systems that are closer to [reaching] human-like [and] animal-like abilities,” he added, describing natural intelligence.

“My research lies at the intersection of computer vision and machine learning, and my overall goal is to create vision systems that are reliable and accessible for everyone,” added Professor Judy Hoffman of the Georgia Institute of Technology.

For the Coexistence of Humans and AI

Day one of the forum closed with a panel discussion in which academics engaged in lively conversations and shared their insights. The panel’s moderator, Youngsang Choi, Vice President of SAIT, introduced topics related to each panelist’s area of expertise. After the discussion, participants were given free rein to ask the panelists questions.

One participant asked the panelists if they believed that it would be possible for AI algorithms to achieve human-level data efficiency in training, to which Professor Antonio Torralba said yes.

“When we think about the data that humans have, it’s not just visual data. They really sense the world through a lot of different mechanisms,” Professor Torralba explained. “Also, humans actually are not passive observers of the world. They are actually interacting with the world and performing all kinds of experiments. I think, in order to achieve [a human-like] level of efficiency, we need to incorporate all of these things and make them really like the main characters of the movie that AI is playing now.”

The panel discussion also offered an opportunity for students majoring in AI-related fields to share their concerns with the experts and receive advice. In the field of AI natural language processing (NLP), for example, the number of parameters is continuously increasing, which means that the costs required to train a model are too. Considering these circumstances, participants discussed which way academic research should be heading.

Professor Bengio concluded the panel discussion by offering some insightful advice to young AI researchers and students. “Don’t be afraid to go in directions that are very different from what has been established as state of the art,” said the professor. “Brain power is the thing that’s really behind innovation and [the] amazing progress that science brings us. So don’t be afraid to try things [and] don’t be afraid to question what has been apparently established for years or decades. That’s how we are all going to make progress.”

Samsung AI Forum Day Two

The Latest AI Research, All in One Place

Day two of the forum was hosted by Samsung Research, Samsung Electronics’ advanced R&D hub, which leads the development of future technologies for its Consumer Electronics and IT & Mobile Communications divisions. Dr. Sebastian Seung, President and Head of Samsung Research, emphasized that “AI is a technology that makes people’s lives better,” and offered an overview of the various AI-related projects that Samsung Research was engaged in, including those related to smartphone cameras, on-device AI, Open Source AI System Software, Machine Translation, and AI technologies for robots. “I’m really looking forward to today’s lectures by leading researchers in AI,” said Dr. Seung, heightening viewers’ expectations.

The day began with a keynote from Professor Leslie Valiant of Harvard University, who offered details on how to augment supervised learning with reasoning. “To make AI work, it takes several components,” Professor Valiant explained. “The first component is identifying which phenomenon or functionality you want to realize.”

Next came lectures delivered by academics who have been actively leading AI research. These include Professor Felix Heide of Princeton University, Research Scientist Been Kim of Google Brain and Professor Max Welling, a research chair in machine learning at the University of Amsterdam and a Distinguished Scientist at Microsoft Research.

AI’s Evolution Into a Tool for Gaining Insights

Day two’s panel discussion saw experts share their opinions on how AI technology will impact people’s lives in the future. The panel’s moderator, Dr. Daniel D. Lee, Executive Vice President and Head of Samsung Research’s Global AI Center, kicked off the discussion with a question.

“The first time AI came into presence, there was a lot of emphasis at that time on logical reasoning,” Dr. Lee explained. “But now, data-driven approaches such as deep neural networks are rising. And what we just heard from Leslie’s talk was [about] how we can actually use the logic [now] in combination with these more advanced neural network techniques. What would be the big advantage of doing that kind of return, in some sense, to logic with neural networks?”

“The idea that both learning and logic are important has been understood for a long time, [albeit separately],” Professor Valiant explained. “We are in a good position because I think the position of learning is now very much advanced. So, we have reason to be confident that there’s a lot of competence that we have as far as learning, and it’s a good basis on which to build logic.”

Researcher Efi Tsamoura of the Samsung AI Center in Cambridge added that “An increasing number of applications for many different areas, from computer vision to natural language processing, are taking advantage of background knowledge in order to build more robust and simpler models. Why is that? It’s because logic provides us with the ability to [complement] missing labels and to use the missing labels in order to train the model.” Tsamoura also pointed out that “An increasing number of researchers from different fields, mostly applied fields, are realizing the potential of logic.”

The discussion also touched on scientific discoveries that have been made with machine learning. “I think it’s worth noting that with the advent of 5G and with 6G coming down the pipe, communication networks have gone from being extremely complicated to super-extremely complicated… and I think the opportunities to optimize and manage the systems to make them even more efficient are vast. So I think there is a great chance to bring machine learning and AI tools to bear on the structure and operation of these communication networks to make them more efficient,” said Gregory Dudek, Head of the Samsung AI Center in Montreal. “We’ve had some very nice success in Montreal in adapting the tools that exist to these relatively new problems for that domain, and [have] actually significantly moved the needle to increase the performance of these systems.”

In order to commercialize machine learning for use in various areas, continuous simulations must be conducted. How then can the gap between simulation results and real-world phenomena be narrowed? Professor Welling shared his thoughts: “[Since simulations do not actually reflect all the complexities of the world,] I think probably the solution is some hybrid solution where you would simulate as much as you can, but you also identify where your system is uncertain about its predictions. And at that point, sort of in an active sense, you are then going to acquire data for that particular problem. So, active sensing might be an interesting solution.”

At the Lightning Talks session, employees from Samsung Research’s Global AI Centers presented some of their latest research including Adaptive Sharpness-Aware Minimization(ASAM), which is a deep learning optimizer developed by Samsung Research, and Named Entity Correction for Automatic Speech Recognition (ASR).

The thoughts and findings that were shared at the Samsung AI Forum indicate that a world in which AI is merged seamlessly with our daily lives may not be that far off. Full replays of both days of the Samsung AI Forum 2021, through which viewers can learn more about the current status of AI technology, its applications, and what the future may hold, can be viewed on the event’s official website and on Samsung Electronics’ YouTube channel.

The second day of the event, hosted by Samsung Research, the advanced R&D hub of the company that leads the development of future technologies for Samsung Electronics’ Consumer Electronics division and IT & Mobile Communications division, facilitated discussion around how industry experts and academics alike can further research into AI technologies that directly impact and enhance the lives of all people. It was livestreamed on Samsung Electronics’ YouTube channel, providing opportunities for researchers and students in the AI field around the world to interact with world-renowned academics through Live Panel Discussions.

AI Forum Day 2: Exploring ‘AI’ in a Human World

Dr. Sebastian Seung, President and Head of Samsung Research, began his opening speech by emphasizing that “Artificial intelligence is revolutionizing every single R&D area of Samsung Research. AI technology was thought impossible, but is now becoming a reality and makes people’s lives better.”

During his speech, which introduced Samsung Research’s various areas of AI research, Dr. Seung explained how on-device AI technology is enabling smartphone cameras to offer users new ways to express their creativity and manage other devices such as TVs and air conditioners. He also provided insights into other applications for AI technology, which includes enabling robot vacuums to automatically create indoor 3D maps, detect obstacles and clean a space accordingly.

Dr. Seung concluded his speech by describing his excitement to see what Samsung’s researchers would be presenting and inviting participants to follow Samsung Research’s Twitter account to learn more about the innovative research that the R&D hub is conducting.

Seung’s speech was followed by presentations from world-renowned AI experts who are leading research in their respective fields.

First, Professor Leslie Valiant of Harvard University, the recipient of the 2010 Turing Award – often referred to as the Nobel Prize of computing – delivered the day’s keynote lecture, entitled “How to Augment Supervised Learning with Reasoning”. Using robust logic as an example, Professor Valiant emphasized that combining supervised learning and reasoning should be a key focus for the next generation of machine learning technology.

Professor Valiant’s keynote was followed by a presentation from Professor Felix Heide of Princeton University. Professor Heide’s presentation, entitled “The Differentiable Camera”, discussed the camera technology that utilizes deep learning to enhance image quality.

Next, research scientist Been Kim of Google Brain delivered a presentation entitled “Interpretability for Skeptical Minds”, in which she shared the latest advancements in interpretable machine learning and proposed directions that this cutting-edge technology should be heading.

The last session of the day was led by Professor Max Welling, a research chair in machine learning at the University of Amsterdam and a Distinguished Scientist at Microsoft Research. During his presentation, entitled “Understanding Matter With Deep Learning”, Professor Welling shared why he is so excited about the scientific breakthroughs that will come from utilizing deep learning in molecular simulation.

Other highlights of Day 2 included a Lightning Talks session, which saw engineers from Samsung Research’s Global AI Center present some of their latest research, and a live panel discussion that was moderated by Dr. Daniel Lee, Executive Vice President and Head of Samsung Research Global AI Center.

In case you missed it, you can watch a full replay of day two of the Samsung AI Forum 2021 by heading to Samsung Electronics’ YouTube channel.

As technology has evolved, it has changed our lives dramatically. It’s truly startling to think just how different life was before the invention of innovations like smartphones, the internet and PCs.

Recently, AI has emerged as a hot topic in this regard based on its potential impact both on technology and on society. Especially with on-device AI¹, AI will be embedded in devices that we use in everyday lives without necessarily connecting to processors in the cloud. To learn more about this exciting subject, Samsung Newsroom met with the head of the Samsung AI Center Cambridge, Dr. Andrew Blake.

Dr. Blake was formerly the Director of both the Alan Turing Institute (which he also helped found) and, before that, Microsoft’s Cambridge Research Laboratory. As a pioneer in the development of the theory and algorithms that make it possible for computers to behave as seeing machines, he explained how Samsung’s AI and hardware innovations will enrich people’s lives in fundamental ways.

Taking IoT to the Next Level

“AI is what is going to breathe life into IoT,” begins Blake.

On-device AI realizes AI functions by processing AI algorithms on the device itself, without necessarily connecting to the cloud, and that is advantageous for privacy and personal information protection, as well as for security.

Unlocking AI-powered smart devices’ true potential will require a combination of two factors: seamlessly connected hardware and an approach to AI that is human-centric above all else.

“One key area is health and fitness – for example, linking exercise, food and mental wellbeing. Another is communication and memories – especially via photography and video. For that, we have to move past the academic world of prototypes working on high-powered computer systems, and get AI working in a leaner fashion – on the everyday devices that people are using.”

The Right Tools

As Blake notes, Samsung’s wide-ranging device portfolio makes it uniquely qualified to deliver this human-centric future for AI.

“This is a great time to be adding new dimensions to Samsung’s AI capabilities, given the company’s leading market position in devices of all sorts,” says Blake. “On-device AI begins with hardware, and this is why working for Samsung is such a fabulous opportunity for AI researchers.”

“Hardware is the channel that moves us beyond simply smart algorithms, to put those algorithms in everyone’s pockets and homes. The big challenge that Cambridge is addressing is moving high-quality embedded AI beyond specialists’ research labs, where people with PhDs in machine learning and in systems work for several months to implement a new embedded system. We envisage a world where advanced tools enable the world’s software developers to move their AI models, simply and effectively, onto Samsung devices, and we are working hard on those tools.”

As Blake explains, on-device AI, in which AI algorithms are processed on a device itself, rather than sent to the cloud, offers significant advantages here by providing a safe and reliable means to protect users’ privacy and data. “We also need to do that in a way that holds the data close, to reassure people that their data is being held safely and privately,” he added.

A Multi-Disciplinary Approach

When it comes to AI, what exactly does the Cambridge AI center want to bring to consumers? The answer to that question is what Samsung has described as a human-centric approach to AI innovation, which Blake describes in further detail.

“Human-centric AI is about homing in on the areas of life that people really care about,” says Blake. “I believe this will require a multi-disciplinary approach. I am not so excited about a future designed solely by engineers. Instead, we need to collaborate with other disciplines, especially design – hardware, user interfaces, and above all, system design – and with human disciplines such as psychology, to achieve a technological future that really helps people live better.”

Taking this multi-disciplinary approach, the Cambridge AI center endeavors to better understand human behavior by exploring areas like communication of emotion, and further expand the boundaries of user-centric communication.²

Drawing from a Diverse Team

Samsung AI Center Cambridge employs a team of experts of various disciplines, and emphasizes collaboration between them.

“We work together a lot as a team,” says Blake. “Our two Program Directors, Maja Pantic and Nic Lane, are world-experts in non-verbal human behavior and in embedded AI, respectively. We also have quite a few senior specialists in machine learning, in machine vision, in networks and devices, and in computing and cognition. We now have a team of very talented people, and new ideas are flowing freely!” says Blake.

As Blake notes, what makes the Cambridge AI center unique is not just its team’s wide-ranging expertise, but its location as well.

“Cambridge is a very special place,” begins Blake. “The university is one of the strongest in the world in research, and that is coupled with an extraordinary culture of research ventures, and a whole constellation of startups in robotics, medicine, AI, self-driving, and many other areas.”

“Being in this environment is important to us for several reasons. It is a stimulating ecosystem and an extraordinary network; it is a rich source of expert talent; it is well connected to the ‘Golden Triangle’ with London and Oxford.”

Of course, in addition to taking full advantage of the benefits that come with its location, the Cambridge center draws strength from its connections to other AI centers in Samsung Research’s global network.

“I am especially pleased to be connected with Samsung’s other AI centers around the world, where I know some of their internationally renowned scientists well,” says Blake. “I believe that, as we begin to work together, we can bring something special to consumers.”

Having more than 40 years of experience working in the field of AI, Blake added, “I was born in the same year as AI – 1956 – the year the Dartmouth conference famously coined the term AI – and I have been studying AI vision for 40 years. I have been lucky to have such an extraordinary career.”

¹ AI that processes information on a device itself, rather than sending that information to the cloud. Because on-device AI does not rely on outside networks, it is regarded as safer and more reliable than cloud-based AI.

² https://internetofbusiness.com/samsung-uk-to-open-new-ai-centre/

Samsung Electronics last month announced its goal to strengthen its leadership in the global system semiconductor industry by 2030 through expanding its proprietary NPU technology development. The company recently delivered an update to this goal at the conference on Computer Vision and Pattern Recognition (CVPR), one of the top academic conferences in computer vision fields.

This update is the company’s development of its On-Device AI lightweight algorithm, introduced at CVPR with a paper titled “Learning to Quantize Deep Networks by Optimizing Quantization Intervals With Task Loss”. On-Device AI technologies directly compute and process data from within the device itself. Over 4 times lighter and 8 times faster than existing algorithms, Samsung’s latest algorithm solution is dramatically improved from previous solutions and has been evaluated to be key to solving potential issues for low-power, high-speed computations.

Streamlining the Deep Learning Process

Samsung Advanced Institute of Technology (SAIT) has announced that they have successfully developed On-Device AI lightweight technology that performs computations 8 times faster than the existing 32-bit deep learning data for servers. By adjusting the data into groups of under 4 bits while maintaining accurate data recognition, this method of deep learning algorithm processing is simultaneously much faster and much more energy efficient than existing solutions.

Samsung’s new On-Device AI processing technology determines the intervals of the significant data that influence overall deep learning performance through ‘learning’. This ‘Quantization¹ Interval Learning (QIL)’ retains data accuracy by re-organizing the data to be presented in bits smaller than their existing size. SAIT ran experiments that successfully demonstrated how the quantization of an in-server deep learning algorithm in 32 bit intervals provided higher accuracy than other existing solutions when computed into levels of less than 4 bits.

When the data of a deep learning computation is presented in bit groups lower than 4 bits, computations of ‘and’ and ‘or’ are allowed, on top of the simpler arithmetic calculations of addition and multiplication. This means that the computation results using the QIL process can achieve the same results as existing processes can while using 1/40 to 1/120 fewer transistors².

As this system therefore requires less hardware and less electricity, it can be mounted directly in-device at the place where the data for an image or fingerprint sensor is being obtained, ahead of transmitting the processed data on to the necessary end points.

The Future of AI Processing and Deep Learning

This technology will help develop Samsung’s system semiconductor capacity as well as strengthening one of the core technologies of the AI era – On-Device AI processing. Differing from AI services that use cloud servers, On-Device AI technologies directly compute data all from within the device itself.

On-Device AI technology can reduce the cost of cloud construction for AI operations since it operates on its own and provides quick and stable performance for use cases such as virtual reality and autonomous driving. Furthermore, On-Device AI technology can save personal biometric information used for device authentication, such as fingerprint, iris and face scans, onto mobile devices safely.

“Ultimately, in the future we will live in a world where all devices and sensor-based technologies are powered by AI,” noted Chang-Kyu Choi, Vice President and head of Computer Vision Lab of SAIT. “Samsung’s On-Device AI technologies are lower-power, higher-speed solutions for deep learning that will pave the way to this future. They are set to expand the memory, processor and sensor market, as well as other next-generation system semiconductor markets.”

A core feature of On-Device AI technology is its ability to compute large amounts of data at a high speed without consuming excessive amounts of electricity. Samsung’s first solution to this end was the Exynos 9 (9820), introduced last year, which featured a proprietary Samsung NPU inside the mobile System on Chip (SoC). This product allows mobile devices to perform AI computations independent of any external cloud server.

Many companies are turning their attention to On-Device AI technology. Samsung Electronics plans to enhance and extend its AI technology leadership by applying this algorithm not only to mobile SoC, but also to memory and sensor solutions in the near future.

Four individuals who played key roles in developing Samsung’s On-Device AI Lightweight Algorithm. From Left to right; Jae-Joon Han, Chang-Young Son, Sang-Il Jung, Chang-Kyu Choi of Samsung Advanced Institute of Technology

1 Quantization is the process of decreasing the number of bits in data by binning the given data into sections of limited number levels, which can be represented in certain bit values and are regarded as having the same value per section

² Transistors are devices that control the flow of current or voltage in a semiconductor by acting as amplifiers or switches