The policy to foster the domestic system semiconductor ecosystem began to operate in earnest in 2019. To strengthen the Korean fabless industry, which holds a global market share of only around 1%, the government has provided support by establishing various measures, such as the 2030 Semiconductor Vision and the Global Star Fabless Development Project.

Significant advancements in the technological capabilities of domestic fabless companies are being made due to fostering policies and changes in the landscape of the industrial ecosystem. We explored the technological capabilities of domestic fabless companies at NextRise 2024, an exhibition event for promising startups.

■ Targeting the market with the world's first active EMI filter... PoC underway with a major corporation

EM Coretech is the world's first company to introduce active EMI filters to the market and holds numerous patents and an IP portfolio related to EMI filters capable of canceling conducted electromagnetic noise.

Although global chip manufacturer Texas Instruments (TI) also commercialized active EMI filters around the same time, it features a non-isolated structure, whereas EM Coretech's product, with its fully magnetically isolated structure, possesses significantly superior characteristics in high-voltage reliability.

Active filters can attenuate electromagnetic waves by independently analyzing incoming waves and emitting an opposite signal. Passive filters, which consist of wound choke coils, have the disadvantages of being large, heavy, and susceptible to overheating. Active filters are smaller, lighter, and have relatively advantageous characteristics regarding heat generation.

"We can reduce CE electromagnetic noise by up to 34dB in the 150kHz to several MHz band," said Shin-ae Ju, Deputy General Manager at EM Coretech. "We are currently conducting reliability tests for application in next-generation products from companies such as Samsung, Hyundai, and LG."

Active EMI filters are expected to be promising in the power electronics field, particularly targeting electric vehicles, solar inverters, and ESS applications.

■ IP Coveted Globally... Gwanak Analog Implements On-Device AI Chip

On this day, Gwanak Analog unveiled a System-on-Chip (SoC) equipped with an on-chip inference accelerator (IX) based on a CNN architecture. Designed based on the Arm Cortex-M4F architecture, Gwanak Analog's GW2110 enables real-time Text-to-Speech (TTS) and keyword recognition functions with low power consumption and low cost.

In particular, as an on-device AI chip that performs AI locally without relying on cloud servers, it is equipped with multiple interfaces, including an analog voltage input channel, to process various sensor information. For example, it is possible to provide a wide range of sensing monitoring, from gas sensors for detecting CO and CH4 to pressure, temperature, and humidity sensors.

Through this, it is expected that the fusion of AI inference accelerators and sensing will enable the implementation and support of various application functions, as well as utilization for data analysis. It is even rumored that the design is so exceptional that global companies have made offers to purchase the IP.

Jeong Yong-sik, Managing Director of Gwanak Analog, stated, “The key point is that there is no DRAM in the single core.” He added that while existing AI processors have no choice but to exchange data with external DRAM for computational processing in multi-cores, Gwanak Analog’s GW2110 is designed with an MCU-based structure without DRAM, enabling AI implementation while achieving a very low price of less than $5 and low power consumption, which is a major selling point.

If on-device AI chips are utilized, ambient sensing information and voice information can be processed without waking up the main processor of home appliances such as TVs, refrigerators, and air conditioners, leading to the operation of the main processor, and fire prediction detection can be implemented in ESS applications based on various sensing information.

■ Edge NPU Chips Replacing NVIDIA GPUs

Mobilint was showcasing two edge NPU chips that provide AI inference performance at its exhibition booth on this day. Introduced the ARIES AI accelerator chip and the new REGULUS AI SoC.

The ARIES AI chip entered full-scale mass production this year and was launched through a collaboration between Semi-Five and Samsung Electronics Foundry. It offers a base AI inference performance of 64 TOPS and can be boosted up to 80 TOPS. Mobilint introduced the ARIES chip as a high-performance AI chip targeting applications such as smart factories, robotics, and intelligent CCTV, stating that it demonstrates AI performance equivalent to the NVIDIA GeForce RTX 4090.

Conversely, as a lighter, lower-power chip than the new REGULUS AI SoC, it targets smaller applications with power consumption of less than 2W. It is reported to have performance equivalent to the NVIDIA Jetson Orin Nano.

The REGULUS AI SoC is scheduled to enter mass production starting at the end of this year and is expected to be sold in the market in earnest from early next year.

Kim Jun-yeop, Team Leader at Mobilint, emphasized, “NVIDIA’s development ecosystem, including software, is so well-established that we are striving to raise the level of support required for UI and embedded implementation in response to it,” adding that “Mobilint’s strength lies in its hardware.”

Mobilint’s chip was able to emphasize its hardware advantages by achieving twice the AI performance of competing products, reducing cost efficiency by half, and significantly lowering power consumption to 15%.