Building 'Vector DB' for Smart AI… Startup That Designs Its Own Semiconductors

[For more diverse corporate information on the startup mentioned in this article, please visit the Unicorn Factory Big Data Platform 'Data Lab'.]

"Even the smartest person cannot perform well without accurate information about their work. The same applies to AI (artificial intelligence). When you ask ChatGPT to 'create our company's production strategy for this year,' it can only provide generic answers because it lacks access to your company's data. This is also why Dinothia is developing a solution that accurately and quickly provides the data needed for AI computation."

Jeong Moo-kyung, CEO of Noh Ti-si-a (CEO), explained as follows regarding the 'C-Horse' solution developed by her company. Its key feature lies in classifying and arranging customer-related data according to meaning and context (vectorization) to efficiently deliver accurate data to AI systems.

"Just as humans frequently search for materials while working, AI must also receive necessary data on demand whenever it executes a command to produce proper results," said Jeong (CEO). "In this process, gathering all the world's data is unnecessary; the core lies in accurately identifying and supplying only the required data by pre-understanding its meaning and context."

As AI technology advances rapidly, the market for vector DB technology developed by Dinothia is also growing quickly. Global market research firm CB Insights added the 'Vector DB' category to its 'AI 100' list, which selected 100 leading AI startups worldwide last year. Dinothia was named among the representative companies in this field alongside German firm Quadrant.

The company is also achieving results in the market. C-Horse was released as a beta version in April and as a full version in September last year, yet it has already generated approximately 3 billion won in revenue within less than a year of launch. Jeong (CEO) stated, "As AI models become more standardized, data plays an even more decisive role in processing results compared to computing power, leading to continuous inquiries from various customers."

Dinothia is simultaneously developing a dedicated device (computer) that maximizes the efficiency of C-Horse. This device will be equipped with its own designed semiconductor, 'VDPU' (Vector Data Processing Unit). It is a semiconductor that enhances efficiency by more than tenfold by complementing the CPU (Central Processing Unit), which handles data searches. This differs from NPU (Neural Processing Unit), which aims to replace GPU (Graphics Processing Unit).

Recently, Dinothia completed 'tape-out,' delivering the final design drawings of VDPU to a foundry. The foundry is TSMC, and production will be carried out using a 12-nanometer process. The company aims to secure its first sample chip by June and begin commercialization in early next year. Dinothia expects that once devices equipped with VDPU are mass-produced, market expansion will accelerate further.

For a startup lacking both talent and capital, developing both software solutions and semiconductors simultaneously was not easy. However, Jeong (CEO) believed that as data utilization becomes increasingly important and data grows exponentially, dedicated semiconductors are necessary. Jeong (CEO) stated, "In the stage of AI application, the proportion of data search and provision computation is growing even faster."

In fact, Jeong (CEO) could have founded a fabless company specializing solely in semiconductor development. Before founding the company, he worked as a semiconductor development engineer at Samsung Electronics' Advanced Institute of Technology, SK Telecom, and Sapion (now merged with Revelion). However, creating a semiconductor-focused enterprise without software was far from his vision.

"To gain market recognition for custom semiconductors, one must fully understand the technology of the software that these semiconductors compute," said Jeong (CEO). "Among various attempts to replace GPUs, Google's TPU (Tensor Processing Unit), which is specialized for AI computation, stands out as achieving the most remarkable results, and this is in the same context."

This year is a significant one for Jeong (CEO). The C-Horse solution is now being supplied to the market in earnest, and the first semiconductor sample has also been produced. Jeong (CEO) stated, "The importance of data is growing even faster than we anticipated when we founded the company in 2023. We will accelerate our efforts to penetrate the solution market and ensure the production of the first semiconductor proceeds without any delays."

[MoneyToday Startup Media Platform Unicorn Factory]