[Edaily Reporter Yun Junghoon ]SamsungElectronics(005930)As Samsung is set to hold its Galaxy Unpacked event in London, UK, on the 22nd (local time) to unveil the Galaxy Z Flip 8, Z Fold 8, and Z Fold 8 Ultra, technology companies that have collaborated on its in-house AP, the “Exynos 2600,” are drawing attention. #Nota Inc. is said to have proven its technical capabilities by supplying AI model optimization technology for the Exynos AP for three consecutive years.

According to the IT industry on the 21st, the Galaxy Z Flip 8 will be released with the Exynos 2600 in select regions, including South Korea and Europe. Nota Inc. is a partner that began collaborating with SamsungElectronics starting with the Exynos 2400 and has supplied optimization technology through the 2500 and 2600 models. Consequently, the success of the new foldable phone is expected to further strengthen Nota Inc.’s position within the value chain.

The Exynos 2600, which will be featured in this device, is a next-generation mobile application processor (AP) manufactured using Samsung Foundry’s 2-nanometer (nm) process. According to the MLPerf mobile benchmark results recently released by SamsungElectronics, it achieved more than double the performance of its predecessor in natural language processing (NLP) and image generation models. The performance of the Neural Processing Unit (NPU)—which is central to generative AI processing—has also been significantly improved, and on-device AI features such as real-time translation, image generation, and AI assistants are expected to be further enhanced.

These hardware performance improvements are the result of Samsung’s process and design technologies. However, no matter how fast the chip becomes, if the AI models running on it are heavy and inefficient, the actual user experience and battery life may suffer. This is where Nota Inc.’s technological expertise comes into play.

Nota Inc.’s proprietary platform, “Netzpresso,” provides technology that streamlines large generative AI models without sacrificing accuracy and automates quantization and computational graph optimization tailored to specific hardware. Simply put, it ensures that the hardware headroom provided by the Exynos 2600 is fully utilized by actual generative AI functions (such as image generation, voice assistants, and real-time translation) without latency and with minimal battery consumption.

Nota Inc.’s technical capabilities are also evident in the case studies it has publicly released. According to the official Netzpresso website, Nota Inc. optimized a healthcare startup’s vision CNN model, “HTCNN,” to run on a low-power microcontroller, achieving a 125-fold speed improvement by reducing inference time from 300 seconds to 2.4 seconds.

In the field of large language models (LLMs), the Upstage Consortium optimized models such as Solar-31B for server and NPU environments, reducing memory usage from 61.8 GB to approximately 19 GB—a reduction of over 60%.

Furthermore, when deploying an LLM with a Mixed Expert (MoE) architecture to an NVIDIA A100 server environment, the number of required GPUs was reduced from four to two, resulting in a 50% reduction in inference costs. In addition, the company has secured various optimization case studies, including achieving an 18-fold speed improvement and increased accuracy simultaneously while porting a vision-language model (VLM) to an NPU, as well as a 6-fold speed improvement in real-time 1080p video processing on edge devices.

However, these figures are not specific to the Exynos 2600 but rather a compilation of various customer and hardware cases—including MCUs, server GPUs, and third-party NPUs—on which Nota Inc. has collaborated; the specific performance improvement when the Exynos 2600 is used on its own has not yet been officially disclosed.

Nota Inc. is also participating in the development of the next-generation version of Samsung’s on-device AI development platform, “Exynos AI Studio.” The plan is to automate the optimization pipeline to help developers port the latest generative AI models to the Exynos platform more quickly. This will enable users to utilize high-performance generative AI features directly on their devices without needing a connection to a cloud server.

An industry insider stated, “We believe that the future competitiveness of smartphones will depend not on the number of AI features, but on how quickly and efficiently AI can be implemented within limited hardware,” adding, “The smartphone market, including the foldable smartphones to be unveiled at this Galaxy Unpacked event, will serve as a stage to demonstrate how efficiently AI can be implemented in a constrained mobile environment.”