Articron Targets 1 Trillion KRW Valuation With In-Memory AI Chips
South Korean startup Articron plans a 2030 IPO backed by Compute-in-Memory SRAM designs that dramatically cut AI power consumption.

South Korean semiconductor startup Articron is targeting a valuation exceeding 1 trillion KRW alongside a stock market debut by 2030, anchored by its on-chip memory designs for artificial intelligence hardware.
Founded three and a half years ago, the venture currently employs roughly 50 people under chief executive Han-ul Jeong. Before joining Yonsei University's electrical and electronic engineering department, Jeong began the enterprise while teaching at Kwangwoon University, building on his previous experience as an SRAM engineer at Samsung Electronics. The startup's moniker blends artificial intelligence with creative association, and its work has received backing from the South Korean Ministry of SMEs and Startups through the Super Gap Startup 1000+ initiative.
According to a report by TheElec, Articron brought in 421 million KRW in revenue last year against a slight net deficit of 21 million KRW. Most of that turnover comes from licensing standard SRAM layouts and proprietary automated design tools. Customers deploy the firm's custom memory components across manufacturing lines run by TSMC and Samsung Electronics, with future process expansions planned for GlobalFoundries and UMC.
Overcoming the AI Memory Wall
While DRAM provides dense bulk capacity using just one transistor per cell, hauling information from external memory packages introduces latency and drains energy. Conventional SRAM avoids external bus delays by keeping inputs on the exact same silicon die as compute engines, though its six-transistor footprint takes up far more physical area. Jeong notes that because machine learning workloads consist largely of continuous additions and multiplications, feeding compute units quickly is essential, making local SRAM capacity a deciding performance factor.
To push efficiency further, Articron is prioritizing Compute-in-Memory (CIM) architecture for neural processing units. Instead of pulling information out of storage cells to execute logic, CIM performs calculations directly inside the memory array. Jeong claims this method preserves workload speed while slashing electrical power requirements by half and reducing total silicon area by 50 percent compared to conventional processor setups.
Roadmap to Public Listing
Articron is initially directing its CIM technology toward convolutional neural networks, which handle computer vision workloads like visual feature extraction and object detection. Future development targets include recurrent neural networks that process sequential speech or text streams, followed by architectures built for on-device large language models.
Monetizing the NPU portfolio will take patience, with commercial revenue from the technology expected to take at least two years. Having accumulated roughly 3 billion KRW in venture capital so far, Articron intends to secure a fresh funding round during the first half of next year, keeping the startup on track for a planned flotation between 2029 and 2030.



