Sunday, September 14, 2025

From Creation to Compliance: An A-to-Z Guide to Copyright Issues in AI-Assisted Fashion Design

 

AI-Designed Clothes: So, Who's the Real Creator?

As generative AI continues to transform the fashion industry, complex copyright questions—particularly around the notion of authorship—are rapidly emerging. In this creative process, prompts play a pivotal role in shaping the human contribution, whether in visual design, music, audiovisual works, or even technical ideas and inventions. This shared reliance on human creative input suggests that a common legal framework may apply across these domains. In what follows, we explore the key legal boundaries and practical considerations that every designer and creator should understand.

Hey there! Have you seen the latest from New York Fashion Week? Brands like Collina Strada are making waves with unique prints and silhouettes created using generative AI. It’s official—AI is no longer just a novelty; it’s becoming a core part of the fashion industry. Students at FIT are using it to analyze trends, and one Hong Kong fashion show even produced over 80 AI-assisted garments.

AI is showing incredible potential in the early stages of brainstorming and concept development. But behind this dazzling technology lie some complex and sensitive legal questions. ‘Who actually owns the copyright to an AI-assisted design?’ and ‘What’s the risk of infringing on the countless existing designs the AI learned from?’ These are fundamental issues we’ll face across all creative fields. Today, we’re going to take a deep dive into this challenging but crucial topic, focusing on legal discussions in the EU and the UK. ๐Ÿ˜Š

 

AI Meets Fashion Design: The Reality and Its Limits

The traditional fashion design process follows a long sequence: trend research, ideation, sketching, and prototyping. Generative AI has become a true ‘game-changer,’ especially in the initial brainstorming and concept development phases. It allows designers to dramatically save time and effort and find new inspiration when facing a creative block.

However, AI isn't a magic bullet. Studies have shown that while AI-generated sketches have high visual quality, they have significant limitations in originality and the ability to reflect a designer's detailed intent.

๐Ÿ’ก An Interesting Paradox!
In one study, the ‘designer input and customization’ aspect of AI performance received the lowest scores. Yet, that’s precisely what designers value the most! This clearly shows that AI is unlikely to ever fully replace a designer's unique artistic sensibility.

 

The Core of Copyright: Was There a ‘Human Creative Choice’?

Let’s get to the most important question: who owns the copyright to an AI-assisted design? The fundamental principle of current law is ‘human-centric.’ This means works created solely by AI cannot receive copyright protection.

The issue arises when a human designer uses AI as a ‘tool.’ In the European Union (EU) and the UK, the standard of ‘Author’s Own Intellectual Creation (AOIC)’ is applied. The key here isn’t the aesthetic quality of the final product, but whether the designer made ‘free and creative choices’ that reflect their personality during the creation process.

A 3-Step Evaluation Process for Copyright

  1. Prompt Curation: The prompt a designer inputs into the AI is itself a result of creative choices. A detailed prompt like, “a red, midi-length, sleeveless A-line dress with ruffle details on the shoulders,” reflects the designer's original thought process.
  2. Output Selection and Modification: The designer’s ‘personal touch’ is added when they ‘select’ a specific design from numerous AI-generated outputs and ‘modify’ it to fit their vision.
  3. Final Completion: The designer’s personality and creativity are fully expressed when they add final details to the selected AI output and bring it to life as a physical garment.

Ultimately, if a human designer's creative intervention can be sufficiently proven throughout these stages, there's a possibility for the AI-assisted design to be protected by copyright.

 

The Complex Dilemma of Third-Party Infringement

While securing copyright for your own design is crucial, the risk of unknowingly infringing on someone else’s copyright is an even bigger concern. Generative AI learns through a process called Text-and-Data Mining (TDM), which scrapes vast amounts of data from the internet. This training data can include copyrighted content, and the process itself can be considered an act of ‘reproduction.’

Copyright infringement is typically determined by three factors: the act of reproduction, a causal link, and substantial similarity. The key is whether “the reproduction of a ‘substantial part’ that constitutes the ‘author's own intellectual creation’” of the original work has occurred.

⚠️ Beware! AI’s ‘Overfitting’ and ‘Memorization’
When an AI model is excessively trained on certain data (overfitting), it may simply regurgitate something nearly identical to its training data instead of creating something new. For instance, there was a case where DALL-E produced strikingly similar images of a red dress—down to the length, neckline, and slit location—when given the same prompt. This is a serious red flag that could lead to unintentional copyright infringement.

To combat this risk, AI companies are implementing ‘AI output filtering technology’. Furthermore, with the EU’s new AI Act, which will require providers to disclose summaries of their training data, it will become easier to assess potential copyright infringement in the future.

 

Legal Defense Strategy: Claiming ‘Transformative Work’

Fortunately, there are ways to navigate these challenges. Even if an AI's initial output is similar to a copyrighted work, a designer can “create a new, non-infringing ‘derivative’ or ‘transformative’ work through sufficient modifications” that adds new meaning or message.

[Case Study] Spain’s ‘Vegap v Mango’ Case

In this case, a court ruled that transforming a copyrighted painting into a digital fashion item was not infringement. It determined the new piece was “‘transformative’ and non-infringing because it provided a new expression, meaning, message, or expanded utility.”

This is an important precedent, showing that designs based on AI-generated content can be considered independent works if they are sufficiently differentiated through the designer's creative intervention.

Therefore, designers using AI should be sure to follow these practical steps:

๐Ÿ“Œ Legal Safeguards for Designers in the AI Era
  1. Systematically Document Your Creative Process: It’s crucial that “the designer's personality and creative freedom are present and documented.” This record of your prompts and modifications is your strongest evidence in a legal dispute.
  2. Use Safe Tools: Remember that copyright exceptions may apply when using AI tools for “non-commercial or private research purposes,” and choose AI tools with clear licensing.
  3. Maximize Your Creative Contribution: Use AI outputs as a ‘first draft’ and focus on transforming them into something truly your own, “differentiated enough that the ‘substantial part’ of the original work is no longer recognizable in the new piece.”
๐Ÿ’ก

AI Fashion Copyright: Key Takeaways

The Author: Only a ‘human’ can be an author. AI is just a tool.
Protection Standard: The key test is whether it’s a human’s ‘Own Intellectual Creation (AOIC)’.
Core Strategy:
‘Transform’ the AI output and ‘Document’ your entire creative process!
Future Value: True value comes from the human's unique perspective, experience, and sensibility in using AI.

Frequently Asked Questions

Q: Is it copyright infringement to ask an AI to design in the style of a specific designer?
A: No. As a general rule, “the mere reproduction of a ‘style’ does not constitute copyright infringement.” However, if you ask it to replicate the unique ‘expressive elements’ of a specific design (e.g., “a small design element with originality”) and the output does so, it could be considered infringement.
Q: Can I claim copyright if I just slightly modify an AI-generated design?
A: The definition of ‘slightly’ is key here. A simple color change or minor detail adjustment may not be enough. To claim your copyright, there needs to be a ‘transformative use’ that is evaluated by the “‘sufficiency’ of the creative choices,” making it substantially different from the original.
Q: What is the controversial ‘Recognizability’ test?
A: It’s a strict new test proposed by an Advocate General at the Court of Justice of the EU. It suggests that if the creative elements of the original work are “‘recognizable’ in the final product, it constitutes infringement.” It has been criticized in the fashion industry for potentially “stifling innovation and creativity,” and has not yet been formally adopted.
Q: How can I know if an AI’s training data includes copyrighted works?
A: While this used to be difficult, regulations like the EU’s AI Act are changing things. The Act will require AI providers to release a ‘sufficiently detailed summary’ of the content used for training. This will bring much-needed transparency and make it easier for users to assess risks.

Conclusion: The Future of Creativity and the Human Role

The copyright debate around AI-assisted fashion design reminds us of a crucial truth: no matter how advanced technology gets, the core of creation remains a uniquely human domain. AI is a powerful tool, but how that tool is used and in which direction it’s guided ultimately depends on the creative choices of a human designer.

In the new paradigm of human-AI collaboration, true value will come from human emotion, experience, and a unique perspective on the world. Amid the infinite possibilities that AI offers, it is still the human touch that elevates an output into a meaningful creation. If you have any more questions, feel free to leave a comment below!

※ Notice ※
This article is primarily based on the paper titled “Generative AI in fashion design creation: a copyright analysis of AI-assisted designs” (Lapatoura et al.), published in the Journal of Intellectual Property Law & Practice.
And this blog post is for general informational purposes only and cannot substitute for legal advice on specific matters. Please be sure to consult with a professional regarding individual legal issues.

์ฐฝ์ž‘์—์„œ ๊ทœ๋ฒ”๊นŒ์ง€: AI ๋ณด์กฐ ํŒจ์…˜ ๋””์ž์ธ์—์„œ์˜ ์ €์ž‘๊ถŒ ์Ÿ์  A to Z ๊ฐ€์ด๋“œ

 

AI๊ฐ€ ๋””์ž์ธํ•œ ์˜ท, ๊ณผ์—ฐ ๋ˆ„๊ตฌ์˜ ์ฐฝ์ž‘๋ฌผ์ผ๊นŒ์š”?

์ƒ์„ฑํ˜• AI๊ฐ€ ํŒจ์…˜ ์‚ฐ์—…์„ ๋น ๋ฅด๊ฒŒ ๋ณ€ํ™”์‹œํ‚ค๋ฉด์„œ, ํŠนํžˆ ‘์ฐฝ์ž‘์˜ ์ฃผ์ฒด’ ๊ฐœ๋…์„ ๋‘˜๋Ÿฌ์‹ผ ๋ณต์žกํ•œ ์ €์ž‘๊ถŒ ๋ฌธ์ œ๊ฐ€ ๋ถ€์ƒํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ด ์ฐฝ์ž‘ ๊ณผ์ •์—์„œ ํ”„๋กฌํ”„ํŠธ๋Š” ์‹œ๊ฐ ๋””์ž์ธ, ์Œ์•…, ์˜์ƒ, ๋‚˜์•„๊ฐ€ ๊ธฐ์ˆ ์  ์•„์ด๋””์–ด๋‚˜ ๋ฐœ๋ช…์— ์ด๋ฅด๊ธฐ๊นŒ์ง€ ์ธ๊ฐ„์˜ ์ฐฝ์˜์  ๊ธฐ์—ฌ๋ฅผ ํ˜•์„ฑํ•˜๋Š” ํ•ต์‹ฌ์ ์ธ ์—ญํ• ์„ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๋ถ„์•ผ ์ „๋ฐ˜์— ๊ฑธ์ณ ์ธ๊ฐ„์˜ ์ฐฝ์ž‘์„ฑ์„ ๊ธฐ์ค€์œผ๋กœ ํ•œ ๊ณตํ†ต๋œ ๋ฒ•์  ํ‹€์ด ์ ์šฉ๋  ์ˆ˜ ์žˆ๋Š” ์ด์œ ๋„ ์—ฌ๊ธฐ์— ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฒˆ ๊ธ€์—์„œ๋Š” ๋””์ž์ด๋„ˆ์™€ ์ฐฝ์ž‘์ž๊ฐ€ ๋ฐ˜๋“œ์‹œ ์•Œ์•„์•ผ ํ•  ํ•ต์‹ฌ ๋ฒ•์  ๊ฒฝ๊ณ„์™€ ์‹ค๋ฌด์  ๊ณ ๋ ค์‚ฌํ•ญ์„ ๊ตฌ์ฒด์ ์œผ๋กœ ์‚ดํŽด๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค.

์•ˆ๋…•ํ•˜์„ธ์š”! ์ตœ๊ทผ ๋‰ด์š• ํŒจ์…˜์œ„ํฌ์—์„œ ‘์ฝœ๋ฆฌ๋‚˜ ์ŠคํŠธ๋ผ๋‹ค’ ๊ฐ™์€ ๋ธŒ๋žœ๋“œ๊ฐ€ ์ƒ์„ฑํ˜• AI๋กœ ๋งŒ๋“  ๋…ํŠนํ•œ ํ”„๋ฆฐํŠธ์™€ ์‹ค๋ฃจ์—ฃ์„ ์„ ๋ณด์ธ ๊ฒƒ, ๋ณด์…จ๋‚˜์š”? ์ด์ œ AI๋Š” ํŒจ์…˜๊ณ„์— ์ƒˆ๋กœ์šด ๋ฐ”๋žŒ์„ ๋„˜์–ด ํ˜์‹ ์˜ ์ค‘์‹ฌ์œผ๋กœ ์ž๋ฆฌ ์žก๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. FIT ๊ฐ™์€ ํŒจ์…˜ ์Šค์ฟจ ํ•™์ƒ๋“ค์€ AI๋กœ ํŠธ๋ Œ๋“œ๋ฅผ ๋ถ„์„ํ•˜๊ณ , ํ™์ฝฉ ํŒจ์…˜์‡ผ์—์„œ๋Š” AI์˜ ๋„์›€์œผ๋กœ 80๋ฒŒ์ด ๋„˜๋Š” ์˜์ƒ์„ ์ œ์ž‘ํ•˜๊ธฐ๋„ ํ–ˆ์ฃ .

์ด์ฒ˜๋Ÿผ AI๋Š” ์•„์ด๋””์–ด ๊ตฌ์ƒ๊ณผ ์ฝ˜์…‰ํŠธ ๊ฐœ๋ฐœ ๋‹จ๊ณ„์—์„œ ๋†€๋ผ์šด ๊ฐ€๋Šฅ์„ฑ์„ ๋ณด์—ฌ์ฃผ๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ์ด ๋ˆˆ๋ถ€์‹  ๊ธฐ์ˆ  ๋ฐœ์ „์˜ ๊ทธ๋ฆผ์ž ์†์—๋Š” ์•„์ฃผ ๋ณต์žกํ•˜๊ณ  ๋ฏผ๊ฐํ•œ ๋ฒ•์  ์งˆ๋ฌธ๋“ค์ด ๋„์‚ฌ๋ฆฌ๊ณ  ์žˆ์–ด์š”. ‘AI๊ฐ€ ์ฐธ์—ฌํ•œ ์ฐฝ์ž‘๋ฌผ์˜ ์ €์ž‘๊ถŒ์€ ๋Œ€์ฒด ๋ˆ„๊ตฌ์˜ ๊ฒƒ์ธ๊ฐ€?’, ‘AI๊ฐ€ ํ•™์Šตํ•œ ์ˆ˜๋งŽ์€ ๊ธฐ์กด ๋””์ž์ธ์„ ์นจํ•ดํ•  ์œ„ํ—˜์€ ์—†์„๊นŒ?’ ํ•˜๋Š” ๊ทผ๋ณธ์ ์ธ ๋ฌธ์ œ๋“ค์ด์ฃ . ์˜ค๋Š˜์€ ์ด ๋ฌด๊ฒ์ง€๋งŒ ๋ฐ˜๋“œ์‹œ ์งš๊ณ  ๋„˜์–ด๊ฐ€์•ผ ํ•  ์ฃผ์ œ๋ฅผ ์œ ๋Ÿฝ์—ฐํ•ฉ๊ณผ ์˜๊ตญ์˜ ๋ฒ•์  ๋…ผ์˜๋ฅผ ์ค‘์‹ฌ์œผ๋กœ ๊นŠ์ด ์žˆ๊ฒŒ ํƒ๊ตฌํ•ด๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค. ๐Ÿ˜Š

 

AI์™€ ํŒจ์…˜ ๋””์ž์ธ์˜ ๋งŒ๋‚จ: ํ˜„์‹ค๊ณผ ํ•œ๊ณ„

์ „ํ†ต์ ์ธ ํŒจ์…˜ ๋””์ž์ธ ๊ณผ์ •์€ ํŠธ๋ Œ๋“œ ์กฐ์‚ฌ, ์•„์ด๋””์–ด ๊ตฌ์ƒ, ์Šค์ผ€์น˜, ํ”„๋กœํ† ํƒ€์ž… ์ œ์ž‘์ด๋ผ๋Š” ๊ธด ์ˆœ์„œ๋ฅผ ๋”ฐ๋ฆ…๋‹ˆ๋‹ค. ์ƒ์„ฑํ˜• AI๋Š” ํŠนํžˆ ์ดˆ๊ธฐ ๋ธŒ๋ ˆ์ธ์Šคํ† ๋ฐ๊ณผ ์ฝ˜์…‰ํŠธ ๊ฐœ๋ฐœ ๋‹จ๊ณ„์—์„œ ๊ทธ์•ผ๋ง๋กœ ‘๊ฒŒ์ž„ ์ฒด์ธ์ €’๊ฐ€ ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋””์ž์ด๋„ˆ๋“ค์€ AI๋ฅผ ํ†ตํ•ด ์‹œ๊ฐ„๊ณผ ๋…ธ๋ ฅ์„ ํš๊ธฐ์ ์œผ๋กœ ๋‹จ์ถ•ํ•˜๊ณ , ์ฐฝ์ž‘์˜ ๋ฒฝ์— ๋ถ€๋”ชํ˜”์„ ๋•Œ ์ƒˆ๋กœ์šด ์˜๊ฐ์„ ์–ป๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค.

ํ•˜์ง€๋งŒ AI๊ฐ€ ๋งŒ๋Šฅ ํ•ด๊ฒฐ์‚ฌ๋Š” ์•„๋‹™๋‹ˆ๋‹ค. ์—ฐ๊ตฌ์— ๋”ฐ๋ฅด๋ฉด AI๊ฐ€ ์ƒ์„ฑํ•œ ์Šค์ผ€์น˜๋Š” ์‹œ๊ฐ์  ์™„์„ฑ๋„๋Š” ๋†’์ง€๋งŒ, ๋…์ฐฝ์„ฑ์ด๋‚˜ ๋””์ž์ด๋„ˆ์˜ ์„ธ๋ฐ€ํ•œ ์˜๋„๋ฅผ ๋ฐ˜์˜ํ•˜๋Š” ๋งž์ถค ์„ค์ • ๋Šฅ๋ ฅ์—์„œ๋Š” ์ƒ๋‹นํ•œ ํ•œ๊ณ„๋ฅผ ๋ณด์˜€์Šต๋‹ˆ๋‹ค.

๐Ÿ’ก ํฅ๋ฏธ๋กœ์šด ์—ญ์„ค!
ํ•œ ์—ฐ๊ตฌ์—์„œ AI ์„ฑ๋Šฅ ํ‰๊ฐ€ ํ•ญ๋ชฉ ์ค‘ ‘๋””์ž์ด๋„ˆ ์ž…๋ ฅ ๋ฐ ๋งž์ถค ์„ค์ •’ ๋ถ€๋ถ„์ด ๊ฐ€์žฅ ๋‚ฎ์€ ์ ์ˆ˜๋ฅผ ๋ฐ›์•˜๋‹ค๊ณ  ํ•ด์š”. ๊ทธ๋Ÿฐ๋ฐ ์ •์ž‘ ๋””์ž์ด๋„ˆ๋“ค์€ ๋ฐ”๋กœ ๊ทธ ๋ถ€๋ถ„์„ ๊ฐ€์žฅ ์ค‘์š”ํ•˜๊ฒŒ ์—ฌ๊ธด๋‹ค๋Š” ์‚ฌ์‹ค! ์ด๋Š” ๊ฒฐ๊ตญ AI๊ฐ€ ๋””์ž์ด๋„ˆ ๊ณ ์œ ์˜ ์„ฌ์„ธํ•œ ๋น„์ „์ด๋‚˜ ์˜ˆ์ˆ ์  ๊ฐ์„ฑ์„ ์™„์ „ํžˆ ๋Œ€์ฒดํ•˜๊ธฐ๋Š” ์–ด๋ ต๋‹ค๋Š” ๊ฒƒ์„ ๋ช…ํ™•ํžˆ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.

 

์ €์ž‘๊ถŒ์˜ ํ•ต์‹ฌ: ‘์ธ๊ฐ„์˜ ์ฐฝ์˜์  ์„ ํƒ’์ด ์žˆ์—ˆ๋Š”๊ฐ€?

๊ฐ€์žฅ ์ค‘์š”ํ•œ ์งˆ๋ฌธ์œผ๋กœ ๋“ค์–ด๊ฐ€ ๋ณด์ฃ . AI ๋ณด์กฐ ๋””์ž์ธ์˜ ์ €์ž‘๊ถŒ์€ ๋ˆ„๊ตฌ์—๊ฒŒ ์žˆ์„๊นŒ์š”? ํ˜„ํ–‰๋ฒ•์˜ ๋Œ€์›์น™์€ ‘์ธ๊ฐ„ ์ค‘์‹ฌ’์ž…๋‹ˆ๋‹ค. ์ฆ‰, AI ๋‹จ๋…์œผ๋กœ ๋งŒ๋“  ์ฐฝ์ž‘๋ฌผ์€ ์ €์ž‘๊ถŒ ๋ณดํ˜ธ๋ฅผ ๋ฐ›์„ ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค.

๋ฌธ์ œ๋Š” ์ธ๊ฐ„ ๋””์ž์ด๋„ˆ๊ฐ€ AI๋ฅผ ‘๋„๊ตฌ’๋กœ ํ™œ์šฉํ–ˆ์„ ๋•Œ์ž…๋‹ˆ๋‹ค. ์œ ๋Ÿฝ์—ฐํ•ฉ(EU)๊ณผ ์˜๊ตญ์—์„œ๋Š” ‘์ €์ž์˜ ๋…์ž์ ์ธ ์ง€์  ์ฐฝ์ž‘๋ฌผ(Author’s Own Intellectual Creation, AOIC)’์ด๋ผ๋Š” ๊ธฐ์ค€์„ ์ ์šฉํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๊ฒฐ๊ณผ๋ฌผ์˜ ์•„๋ฆ„๋‹ค์›€์ด ์•„๋‹ˆ๋ผ, ์ฐฝ์ž‘ ๊ณผ์ •์—์„œ ๋””์ž์ด๋„ˆ๊ฐ€ ์ž์‹ ์˜ ๊ฐœ์„ฑ์„ ๋“œ๋Ÿฌ๋‚ด๋Š” ‘์ž์œ ๋กญ๊ณ  ์ฐฝ์˜์ ์ธ ์„ ํƒ’์„ ํ–ˆ๋Š”์ง€ ์—ฌ๋ถ€๋ฅผ ๋”ฐ์ง€๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

์ €์ž‘๊ถŒ ํŒ๋‹จ์„ ์œ„ํ•œ 3๋‹จ๊ณ„ ํ‰๊ฐ€ ๊ณผ์ •

  1. ํ”„๋กฌํ”„ํŠธ ํ๋ ˆ์ด์…˜: ๋””์ž์ด๋„ˆ๊ฐ€ AI์— ์ž…๋ ฅํ•˜๋Š” ํ”„๋กฌํ”„ํŠธ ์ž์ฒด๊ฐ€ ์ฐฝ์˜์  ์„ ํƒ์˜ ๊ฒฐ๊ณผ๋ฌผ์ž…๋‹ˆ๋‹ค. “๋นจ๊ฐ„์ƒ‰, ๋ฏธ๋”” ๊ธธ์ด, ์†Œ๋งค ์—†๋Š” A๋ผ์ธ ๋“œ๋ ˆ์Šค, ์–ด๊นจ์— ๋Ÿฌํ”Œ ์žฅ์‹”์ฒ˜๋Ÿผ ๊ตฌ์ฒด์ ์ด๊ณ  ์ƒ์„ธํ•œ ํ”„๋กฌํ”„ํŠธ๋Š” ๋””์ž์ด๋„ˆ์˜ ๋…์ฐฝ์  ์‚ฌ๊ณ ๋ฅผ ๋ฐ˜์˜ํ•ฉ๋‹ˆ๋‹ค.
  2. ๊ฒฐ๊ณผ๋ฌผ ์„ ๋ณ„ ๋ฐ ์ˆ˜์ •: AI๊ฐ€ ์ƒ์„ฑํ•œ ์ˆ˜๋งŽ์€ ๊ฒฐ๊ณผ๋ฌผ ์ค‘ ํŠน์ • ๋””์ž์ธ์„ ‘์„ ํƒ’ํ•˜๊ณ , ์ž์‹ ์˜ ๋น„์ „์— ๋งž๊ฒŒ ‘์ˆ˜์ •’ํ•˜๋Š” ๊ณผ์ •์—์„œ ๋””์ž์ด๋„ˆ์˜ ‘๊ฐœ์ธ์ ์ธ ์†๊ธธ(personal touch)’์ด ๋”ํ•ด์ง‘๋‹ˆ๋‹ค.
  3. ์ตœ์ข… ์™„์„ฑ: ์„ ํƒ๋œ AI ๊ฒฐ๊ณผ๋ฌผ์„ ๋ฐ”ํƒ•์œผ๋กœ ์ตœ์ข… ๋””ํ…Œ์ผ์„ ์ถ”๊ฐ€ํ•˜๊ณ  ์‹ค์ œ ์˜๋ฅ˜๋กœ ์™„์„ฑํ•˜๋Š” ๊ณผ์ •์—์„œ ๋””์ž์ด๋„ˆ์˜ ๊ฐœ์„ฑ๊ณผ ์ฐฝ์˜์„ฑ์ด ์ข…ํ•ฉ์ ์œผ๋กœ ๋ฐœํ˜„๋ฉ๋‹ˆ๋‹ค.

๊ฒฐ๊ตญ ์ด ๋ชจ๋“  ๊ณผ์ •์—์„œ ์ธ๊ฐ„ ๋””์ž์ด๋„ˆ์˜ ์ฐฝ์˜์  ๊ฐœ์ž…์ด ์ถฉ๋ถ„ํžˆ ์ž…์ฆ๋œ๋‹ค๋ฉด, AI ๋ณด์กฐ ๋””์ž์ธ๋„ ์ €์ž‘๊ถŒ ๋ณดํ˜ธ๋ฅผ ๋ฐ›์„ ๊ฐ€๋Šฅ์„ฑ์ด ์—ด๋ฆฌ๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค.

 

์ œ3์ž ์ €์ž‘๊ถŒ ์นจํ•ด๋ผ๋Š” ๋ณต์žกํ•œ ๋”œ๋ ˆ๋งˆ

๋‚ด ๋””์ž์ธ์˜ ์ €์ž‘๊ถŒ์„ ์ธ์ •๋ฐ›๋Š” ๊ฒƒ๋„ ์ค‘์š”ํ•˜์ง€๋งŒ, ๋‚˜๋„ ๋ชจ๋ฅด๊ฒŒ ๋‹ค๋ฅธ ์‚ฌ๋žŒ์˜ ์ €์ž‘๊ถŒ์„ ์นจํ•ดํ•  ์œ„ํ—˜์€ ๋” ํฐ ๋ฌธ์ œ์ž…๋‹ˆ๋‹ค. ์ƒ์„ฑํ˜• AI๋Š” ํ…์ŠคํŠธ ๋ฐ ๋ฐ์ดํ„ฐ ๋งˆ์ด๋‹(Text-and-Data Mining, TDM) ๊ธฐ์ˆ ์„ ํ†ตํ•ด ์ธํ„ฐ๋„ท์˜ ๋ฐฉ๋Œ€ํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ํ•™์Šตํ•˜๋Š”๋ฐ, ์ด ๊ณผ์ •์—์„œ ์ €์ž‘๊ถŒ์ด ์žˆ๋Š” ์ฝ˜ํ…์ธ ๊ฐ€ ๋ฌด๋‹จ์œผ๋กœ ‘๋ณต์ œ’๋  ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.

์ €์ž‘๊ถŒ ์นจํ•ด๋Š” ๋ณดํ†ต ‘๋ณต์ œ ํ–‰์œ„’, ‘์ธ๊ณผ ๊ด€๊ณ„’, ‘์‹ค์งˆ์  ์œ ์‚ฌ์„ฑ’ ์„ธ ๊ฐ€์ง€๋ฅผ ๋”ฐ์ ธ ํŒ๋‹จํ•ฉ๋‹ˆ๋‹ค. ํŠนํžˆ ์›๋ณธ ์ €์ž‘๋ฌผ์˜ “‘์ €์ž์˜ ๋…์ž์ ์ธ ์ง€์  ์ฐฝ์ž‘๋ฌผ’์„ ๊ตฌ์„ฑํ•˜๋Š” ‘์‹ค์งˆ์ ์ธ ๋ถ€๋ถ„’์ด ๋ณต์ œ”๋˜์—ˆ๋Š”์ง€๊ฐ€ ํ•ต์‹ฌ์ž…๋‹ˆ๋‹ค.

⚠️ ์ฃผ์˜! AI์˜ ‘๊ณผ์ ํ•ฉ’๊ณผ ‘๊ธฐ์–ต’ ํ˜„์ƒ
AI ๋ชจ๋ธ์ด ํŠน์ • ๋ฐ์ดํ„ฐ๋ฅผ ๊ณผ๋„ํ•˜๊ฒŒ ํ•™์Šตํ•˜๋ฉด(๊ณผ์ ํ•ฉ), ์ƒˆ๋กœ์šด ์ฐฝ์ž‘ ๋Œ€์‹  ํ•™์Šต ๋ฐ์ดํ„ฐ์™€ ๊ฑฐ์˜ ๋˜‘๊ฐ™์€ ๊ฒฐ๊ณผ๋ฌผ์„ ๋ฑ‰์–ด๋‚ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ์‹ค์ œ๋กœ DALL-E์—๊ฒŒ ๋™์ผํ•œ ํ”„๋กฌํ”„ํŠธ๋ฅผ ์ž…๋ ฅํ–ˆ์„ ๋•Œ, ๋“œ๋ ˆ์Šค ๊ธธ์ด, ๋„คํฌ๋ผ์ธ, ์Šฌ๋ฆฟ ์œ„์น˜๊นŒ์ง€ ๋†€๋ž๋„๋ก ์œ ์‚ฌํ•œ ์ด๋ฏธ์ง€๊ฐ€ ์ƒ์„ฑ๋œ ์‚ฌ๋ก€๊ฐ€ ์žˆ์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ์˜๋„์น˜ ์•Š์€ ์ €์ž‘๊ถŒ ์นจํ•ด๋กœ ์ด์–ด์งˆ ์ˆ˜ ์žˆ๋Š” ์‹ฌ๊ฐํ•œ ์œ„ํ—˜ ์‹ ํ˜ธ์ž…๋‹ˆ๋‹ค.

์ด๋Ÿฌํ•œ ์œ„ํ—˜ ๋•Œ๋ฌธ์— AI ๊ธฐ์—…๋“ค์€ ์นจํ•ด ๊ฐ€๋Šฅ์„ฑ์ด ์žˆ๋Š” ๊ฒฐ๊ณผ๋ฌผ์„ ๊ฑธ๋Ÿฌ๋‚ด๋Š” ‘AI ์ถœ๋ ฅ ํ•„ํ„ฐ๋ง ๊ธฐ์ˆ ’์„ ๋„์ž…ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ์•ž์œผ๋กœ EU์˜ AI ๋ฒ•(AI Act)์— ๋”ฐ๋ผ ํ•™์Šต ๋ฐ์ดํ„ฐ ์š”์•ฝ์„ ๊ณต๊ฐœํ•  ์˜๋ฌด๊ฐ€ ์ƒ๊ธฐ๋ฉด ์ €์ž‘๊ถŒ ์นจํ•ด ํŒ๋‹จ์ด ๋” ์šฉ์ดํ•ด์งˆ ์ „๋ง์ž…๋‹ˆ๋‹ค.

 

๋ฒ•์  ๋ฐฉ์–ด ์ „๋žต: ‘๋ณ€ํ˜•์  ์ž‘ํ’ˆ’์œผ๋กœ ์ธ์ •๋ฐ›๊ธฐ

๋‹คํ–‰ํžˆ ํ•ด๊ฒฐ์ฑ…์ด ์—†๋Š” ๊ฒƒ์€ ์•„๋‹™๋‹ˆ๋‹ค. ์„ค๋ น AI์˜ ์ดˆ๊ธฐ ๊ฒฐ๊ณผ๋ฌผ์ด ๋‹ค๋ฅธ ์ €์ž‘๋ฌผ๊ณผ ์œ ์‚ฌํ•˜๋”๋ผ๋„, ์ธ๊ฐ„ ๋””์ž์ด๋„ˆ๊ฐ€ ๊ทธ ์œ„์— “์ถฉ๋ถ„ํ•œ ์ˆ˜์ •์„ ๊ฐ€ํ•˜๋ฉด ์ƒˆ๋กœ์šด ๋น„์นจํ•ด์ ์ธ ‘ํŒŒ์ƒ’ ๋˜๋Š” ‘๋ณ€ํ˜•์ (transformative)’ ์ž‘ํ’ˆ”์ด ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

[์‚ฌ๋ก€] ์ŠคํŽ˜์ธ ๋ฒ•์›์˜ ‘Vegap v Mango’ ์‚ฌ๊ฑด

์ด ์‚ฌ๊ฑด์—์„œ ๋ฒ•์›์€ ์ €์ž‘๊ถŒ์ด ์žˆ๋Š” ๊ทธ๋ฆผ์„ ๋””์ง€ํ„ธ ํŒจ์…˜ ์•„์ดํ…œ์œผ๋กœ ๋ณ€ํ˜•ํ•œ ์ž‘ํ’ˆ์— ๋Œ€ํ•ด, ๋‹จ์ˆœํžˆ ์ด๋ฏธ์ง€๋ฅผ ์˜ฎ๊ธด ๊ฒƒ์ด ์•„๋‹ˆ๋ผ “์ƒˆ๋กœ์šด ํ‘œํ˜„, ์˜๋ฏธ, ๋ฉ”์‹œ์ง€๋ฅผ ์ œ๊ณตํ•˜๊ฑฐ๋‚˜ ์œ ์šฉ์„ฑ์„ ํ™•์žฅํ•˜์—ฌ ‘๋ณ€ํ˜•์ ’์ด๋ฉฐ ๋น„์นจํ•ด์ ”์ด๋ผ๊ณ  ํŒ๋‹จํ–ˆ์Šต๋‹ˆ๋‹ค.

์ด๋Š” AI ์ƒ์„ฑ ๊ฒฐ๊ณผ๋ฌผ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ๋””์ž์ธ์ด๋ผ๋„ ๋””์ž์ด๋„ˆ์˜ ์ฐฝ์˜์  ๊ฐœ์ž…์„ ํ†ตํ•ด ์ถฉ๋ถ„ํžˆ ์ฐจ๋ณ„ํ™”๋œ๋‹ค๋ฉด ๋…๋ฆฝ์ ์ธ ์ €์ž‘๋ฌผ๋กœ ์ธ์ •๋ฐ›์„ ์ˆ˜ ์žˆ์Œ์„ ๋ณด์—ฌ์ฃผ๋Š” ์ค‘์š”ํ•œ ์„ ๋ก€์ž…๋‹ˆ๋‹ค.

๋”ฐ๋ผ์„œ ๋””์ž์ด๋„ˆ๋Š” AI๋ฅผ ํ™œ์šฉํ•  ๋•Œ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์‹ค๋ฌด์  ๋Œ€์‘ ๋ฐฉ์•ˆ์„ ๋ฐ˜๋“œ์‹œ ์ˆ™์ง€ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.

๐Ÿ“Œ AI ์‹œ๋Œ€ ๋””์ž์ด๋„ˆ๋ฅผ ์œ„ํ•œ ๋ฒ•์  ์•ˆ์ „์žฅ์น˜
  1. ์ฐฝ์ž‘ ๊ณผ์ •์˜ ์ฒด๊ณ„์  ๋ฌธ์„œํ™”: ํ”„๋กฌํ”„ํŠธ ์„ค๊ณ„๋ถ€ํ„ฐ ์ตœ์ข… ์ˆ˜์ •๊นŒ์ง€, “๋””์ž์ด๋„ˆ์˜ ๊ฐœ์„ฑ๊ณผ ์ฐฝ์˜์  ์ž์œ ๊ฐ€ ์กด์žฌํ•˜๊ณ  ๋ฌธ์„œํ™””๋˜๋Š” ๊ฒƒ์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ์ด๊ฒƒ์ด ๋ฒ•์  ๋ถ„์Ÿ ์‹œ ๊ฐ€์žฅ ๊ฐ•๋ ฅํ•œ ์ฆ๊ฑฐ๊ฐ€ ๋ฉ๋‹ˆ๋‹ค.
  2. ์•ˆ์ „ํ•œ ๋„๊ตฌ ํ™œ์šฉ: “๋น„์ƒ์—…์  ์—ฐ๊ตฌ ๋˜๋Š” ์‚ฌ์  ์—ฐ๊ตฌ ๋ชฉ์ ”์œผ๋กœ AI๋ฅผ ์‹คํ—˜ํ•˜๋Š” ๊ฒฝ์šฐ ์ €์ž‘๊ถŒ ์˜ˆ์™ธ๊ฐ€ ํ—ˆ์šฉ๋  ์ˆ˜ ์žˆ์Œ์„ ๊ธฐ์–ตํ•˜๊ณ , ๋ผ์ด์„ ์Šค๊ฐ€ ๋ช…ํ™•ํ•œ AI ๋„๊ตฌ๋ฅผ ์„ ํƒํ•˜์„ธ์š”.
  3. ์ฐฝ์˜์  ๊ธฐ์—ฌ๋„ ๊ทน๋Œ€ํ™”: AI ๊ฒฐ๊ณผ๋ฌผ์„ ‘์ดˆ์•ˆ’์œผ๋กœ ์‚ผ๊ณ , ์›๋ณธ์˜ “‘์‹ค์งˆ์ ์ธ ๋ถ€๋ถ„’์„ ์ƒˆ๋กœ์šด ์ž‘ํ’ˆ์—์„œ ์ฐพ์•„๋ณผ ์ˆ˜ ์—†์„ ์ •๋„๋กœ ์ถฉ๋ถ„ํžˆ ์ฐจ๋ณ„ํ™””ํ•˜์—ฌ ์ž์‹ ๋งŒ์˜ ์ž‘ํ’ˆ์œผ๋กœ ์žฌ์ฐฝ์กฐํ•˜๋Š” ๋ฐ ์ง‘์ค‘ํ•˜์„ธ์š”.
๐Ÿ’ก

AI ํŒจ์…˜ ์ €์ž‘๊ถŒ ํ•ต์‹ฌ ์š”์•ฝ

์ €์ž‘๊ถŒ ์ฃผ์ฒด: ์˜ค์ง ‘์ธ๊ฐ„’๋งŒ์ด ์ €์ž‘์ž๊ฐ€ ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. AI๋Š” ๋„๊ตฌ์ผ ๋ฟ์ž…๋‹ˆ๋‹ค.
๋ณดํ˜ธ ๊ธฐ์ค€: ์ธ๊ฐ„์˜ ‘๋…์ž์  ์ง€์  ์ฐฝ์ž‘(AOIC)’์ด ์žˆ์—ˆ๋Š”์ง€๊ฐ€ ํ•ต์‹ฌ ํŒ๋‹จ ๊ธฐ์ค€์ž…๋‹ˆ๋‹ค.
ํ•ต์‹ฌ ์ „๋žต:
AI ๊ฒฐ๊ณผ๋ฌผ์„ ‘๋ณ€ํ˜•(Transform)’ํ•˜๊ณ  ๋ชจ๋“  ์ฐฝ์ž‘ ๊ณผ์ •์„ ‘๊ธฐ๋ก’ํ•˜์„ธ์š”!
๋ฏธ๋ž˜ ๊ฐ€์น˜: ์ง„์ •ํ•œ ๊ฐ€์น˜๋Š” AI๋ฅผ ํ™œ์šฉํ•˜๋Š” ์ธ๊ฐ„์˜ ๊ฐ์„ฑ, ๊ฒฝํ—˜, ๋…ํŠนํ•œ ๊ด€์ ์—์„œ ๋‚˜์˜ต๋‹ˆ๋‹ค.

์ž์ฃผ ๋ฌป๋Š” ์งˆ๋ฌธ

Q: AI์—๊ฒŒ ํŠน์ • ๋””์ž์ด๋„ˆ ์Šคํƒ€์ผ๋กœ ๋””์ž์ธํ•ด๋‹ฌ๋ผ๊ณ  ์š”์ฒญํ•˜๋Š” ๊ฒƒ์€ ์ €์ž‘๊ถŒ ์นจํ•ด์ธ๊ฐ€์š”?
A: ์•„๋‹™๋‹ˆ๋‹ค. “‘์Šคํƒ€์ผ’์˜ ๋ณต์ œ๋งŒ์œผ๋กœ๋Š” ์ €์ž‘๊ถŒ ์นจํ•ด๊ฐ€ ์„ฑ๋ฆฝ๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค.” ํ•˜์ง€๋งŒ ํŠน์ • ๋””์ž์ธ์˜ ๋…์ฐฝ์ ์ธ ‘ํ‘œํ˜„ ์š”์†Œ’(์˜ˆ: “๋…์ฐฝ์„ฑ์ด ์žˆ๋Š” ์ž‘์€ ๋””์ž์ธ ์š”์†Œ”)๋ฅผ ๊ทธ๋Œ€๋กœ ๋”ฐ๋ผ ํ•˜๋„๋ก ์š”์ฒญํ•˜๊ณ  ๊ฒฐ๊ณผ๋ฌผ์ด ๊ทธ๋ ‡๊ฒŒ ๋‚˜์˜จ๋‹ค๋ฉด ์นจํ•ด์˜ ์†Œ์ง€๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
Q: AI๊ฐ€ ๋งŒ๋“  ๋””์ž์ธ์„ ์ œ๊ฐ€ ์กฐ๊ธˆ๋งŒ ์ˆ˜์ •ํ•ด์„œ ์‚ฌ์šฉํ•ด๋„ ์ €์ž‘๊ถŒ์„ ์ฃผ์žฅํ•  ์ˆ˜ ์žˆ๋‚˜์š”?
A: ‘์กฐ๊ธˆ’์˜ ๊ธฐ์ค€์ด ์ค‘์š”ํ•ฉ๋‹ˆ๋‹ค. ๋‹จ์ˆœํ•œ ์ƒ‰์ƒ ๋ณ€๊ฒฝ์ด๋‚˜ ๋ฏธ๋ฏธํ•œ ๋””ํ…Œ์ผ ์ˆ˜์ •๋งŒ์œผ๋กœ๋Š” ๋ถ€์กฑํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. “์ฐฝ์˜์  ์„ ํƒ์˜ ‘์ถฉ๋ถ„์„ฑ’”์„ ํ‰๊ฐ€ํ•˜๋ฉฐ, ์›๋ณธ๊ณผ ์‹ค์งˆ์ ์œผ๋กœ ๊ตฌ๋ณ„๋  ๋งŒํผ์˜ ‘๋ณ€ํ˜•์  ์‚ฌ์šฉ’์ด ์ด๋ฃจ์–ด์ ธ์•ผ ์ž์‹ ์˜ ์ €์ž‘๊ถŒ์„ ์ฃผ์žฅํ•˜๊ธฐ ์œ ๋ฆฌํ•ฉ๋‹ˆ๋‹ค.
Q: ๋…ผ๋ž€์ด ๋˜๋Š” ‘์ธ์ง€ ๊ฐ€๋Šฅ์„ฑ(Recognizability)’ ํ…Œ์ŠคํŠธ๋ž€ ๋ฌด์—‡์ธ๊ฐ€์š”?
A: ์œ ๋Ÿฝ์‚ฌ๋ฒ•์žฌํŒ์†Œ์˜ ํ•œ ๋ฒ•๋ฌด๊ด€์ด ์ œ์•ˆํ•œ ์ƒˆ๋กœ์šด ๊ธฐ์ค€์œผ๋กœ, ์ตœ์ข… ๊ฒฐ๊ณผ๋ฌผ์—์„œ ๋ณต์ œ๋œ ์›๋ณธ์˜ ์ฐฝ์˜์  ์š”์†Œ๊ฐ€ “‘์ธ์ง€ ๊ฐ€๋Šฅ’ํ•˜๋ฉด ์นจํ•ด๊ฐ€ ์„ฑ๋ฆฝ”๋œ๋‹ค๋Š” ๋‹ค์†Œ ์—„๊ฒฉํ•œ ๊ธฐ์ค€์ž…๋‹ˆ๋‹ค. ํŒจ์…˜ ์‚ฐ์—…๊ณ„์—์„œ๋Š” “ํ˜์‹ ๊ณผ ์ฐฝ์˜์„ฑ์„ ์ €ํ•ดํ•  ์ˆ˜ ์žˆ๋‹ค”๋Š” ๋น„ํŒ์„ ๋ฐ›๊ณ  ์žˆ์–ด ์•„์ง ๊ณต์‹ ์ฑ„ํƒ๋˜์ง€๋Š” ์•Š์•˜์Šต๋‹ˆ๋‹ค.
Q: AI๊ฐ€ ํ•™์Šตํ•œ ๋ฐ์ดํ„ฐ์— ์ €์ž‘๋ฌผ์ด ์žˆ๋Š”์ง€ ์ œ๊ฐ€ ์–ด๋–ป๊ฒŒ ์•Œ ์ˆ˜ ์žˆ๋‚˜์š”?
A: ๊ณผ๊ฑฐ์—๋Š” ์•Œ๊ธฐ ์–ด๋ ค์› ์ง€๋งŒ, EU์˜ AI ๋ฒ•(AI Act)์— ๋”ฐ๋ผ AI ์ œ๊ณต์ž๋Š” ํ›ˆ๋ จ์— ์‚ฌ์šฉ๋œ ์ฝ˜ํ…์ธ ์— ๋Œ€ํ•œ ‘์ถฉ๋ถ„ํžˆ ์ƒ์„ธํ•œ ์š”์•ฝ’์„ ๊ณต๊ฐœํ•  ์˜๋ฌด๊ฐ€ ์ƒ๊ฒผ์Šต๋‹ˆ๋‹ค. ์•ž์œผ๋กœ๋Š” ํˆฌ๋ช…์„ฑ์ด ๋” ๋†’์•„์ ธ ์‚ฌ์šฉ์ž๊ฐ€ ์ด ๋ถ€๋ถ„์„ ํŒ๋‹จํ•˜๊ธฐ ๋” ์‰ฌ์›Œ์งˆ ์ „๋ง์ž…๋‹ˆ๋‹ค.

๊ฒฐ๋ก : ์ฐฝ์ž‘์˜ ๋ฏธ๋ž˜์™€ ์ธ๊ฐ„์˜ ์—ญํ• 

AI ๋ณด์กฐ ํŒจ์…˜ ๋””์ž์ธ์˜ ์ €์ž‘๊ถŒ ์Ÿ์ ์€ ์šฐ๋ฆฌ์—๊ฒŒ ์ค‘์š”ํ•œ ์‚ฌ์‹ค์„ ์ผ๊นจ์›Œ์ค๋‹ˆ๋‹ค. ๊ธฐ์ˆ ์ด ์•„๋ฌด๋ฆฌ ๋ฐœ์ „ํ•ด๋„ ์ฐฝ์ž‘์˜ ํ•ต์‹ฌ์€ ์—ฌ์ „ํžˆ ์ธ๊ฐ„์˜ ๊ณ ์œ ํ•œ ์˜์—ญ์ด๋ผ๋Š” ์ ์ž…๋‹ˆ๋‹ค. AI๋Š” ๊ฐ•๋ ฅํ•œ ๋„๊ตฌ์ด์ง€๋งŒ, ๊ทธ ๋„๊ตฌ๋ฅผ ์–ด๋–ป๊ฒŒ ์‚ฌ์šฉํ•˜๊ณ  ์–ด๋–ค ๋ฐฉํ–ฅ์œผ๋กœ ์ด๋Œ์ง€๋Š” ๊ฒฐ๊ตญ ์ธ๊ฐ„ ๋””์ž์ด๋„ˆ์˜ ์ฐฝ์˜์  ์„ ํƒ์— ๋‹ฌ๋ ค ์žˆ์Šต๋‹ˆ๋‹ค.

๊ถ๊ทน์ ์œผ๋กœ AI์™€ ์ธ๊ฐ„์ด ํ˜‘๋ ฅํ•˜๋Š” ์ƒˆ๋กœ์šด ์ฐฝ์ž‘ ํŒจ๋Ÿฌ๋‹ค์ž„์—์„œ ์ง„์ •ํ•œ ๊ฐ€์น˜๋Š” ์ธ๊ฐ„์˜ ๊ฐ์„ฑ, ๊ฒฝํ—˜, ๊ทธ๋ฆฌ๊ณ  ์„ธ์ƒ์„ ๋ฐ”๋ผ๋ณด๋Š” ๋…ํŠนํ•œ ๊ด€์ ์—์„œ ๋‚˜์˜ฌ ๊ฒƒ์ž…๋‹ˆ๋‹ค. AI๊ฐ€ ์ œ๊ณตํ•˜๋Š” ๋ฌดํ•œํ•œ ๊ฐ€๋Šฅ์„ฑ ์†์—์„œ๋„, ๊ทธ๊ฒƒ์„ ์˜๋ฏธ ์žˆ๋Š” ์ฐฝ์ž‘๋ฌผ๋กœ ์Šนํ™”์‹œํ‚ค๋Š” ๊ฒƒ์€ ์—ฌ์ „ํžˆ ์šฐ๋ฆฌ ์ธ๊ฐ„๋งŒ์ด ํ•  ์ˆ˜ ์žˆ๋Š” ๊ณ ์œ ํ•œ ์—ญํ• ์ž…๋‹ˆ๋‹ค. ๋” ๊ถ๊ธˆํ•œ ์ ์ด ์žˆ๋‹ค๋ฉด ๋Œ“๊ธ€๋กœ ๋ฌผ์–ด๋ด์ฃผ์„ธ์š”!

※ ๊ณ ์ง€ (Notice) ※
๋ณธ ๊ธ€์€ Journal of Intellectual Property Law & Practice ๊ฒŒ์žฌ ๋…ผ๋ฌธ "Generative AI in fashion design creation: a copyright analysis of AI-assisted designs"(Lapatoura et al.)๋ฅผ ์ฃผ์š” ์ฐธ๊ณ ์ž๋ฃŒ๋กœ ํ•˜์—ฌ ์ž‘์„ฑ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
๋˜ํ•œ ๋ณธ ๊ธ€์€ ์ผ๋ฐ˜์ ์ธ ์ •๋ณด ์ œ๊ณต์„ ๋ชฉ์ ์œผ๋กœ ํ•˜๋ฉฐ, ํŠน์ • ์‚ฌ์•ˆ์— ๋Œ€ํ•œ ๋ฒ•๋ฅ ์  ์ž๋ฌธ์„ ๋Œ€์ฒดํ•  ์ˆ˜ ์—†์Šต๋‹ˆ๋‹ค. ๊ฐœ๋ณ„์ ์ธ ๋ฒ•๋ฅ  ๋ฌธ์ œ์— ๋Œ€ํ•ด์„œ๋Š” ๋ฐ˜๋“œ์‹œ ์ „๋ฌธ๊ฐ€์™€ ์ƒ๋‹ดํ•˜์‹œ๊ธฐ ๋ฐ”๋ž๋‹ˆ๋‹ค.

Saturday, September 13, 2025

The JSR Takeover and the New Era of Tech Nationalism: A Photoresist Deep Dive

 

How did the liquid in this tiny bottle become the key to semiconductor dominance? We take a deep dive into the unsung hero of chip manufacturing, ‘photoresist,’ and uncover the secrets of how Japanese companies captured 90% of the global market—from its origins to its geopolitical significance.

Hey everyone! Today, we’re talking about a chemical that’s absolutely essential for making semiconductors, but its name might be a little unfamiliar: ‘photoresist.’ Ever heard of it? It’s a fascinating topic because Japanese companies control nearly 90% of this market. That’s almost a complete monopoly.

So today, we’re going to dig deep into how photoresist technology evolved and how Japanese firms reached their current position, exploring everything from the history and strategy to the geopolitical context. I’ll break it all down so you can grasp this complex story quickly and thoroughly! ๐Ÿ˜Š

 

So, What Exactly Is Photoresist? ๐Ÿค”

Simply put, it’s a light-sensitive liquid used to etch microscopic circuit patterns onto a semiconductor wafer. It reacts to light, much like the film used in photography. Without it, even the most expensive, state-of-the-art lithography equipment would be nothing more than a giant paperweight. Think of it this way: no matter how great your printer is, it’s useless without ink. Photoresist acts as that crucial ink or the stencil for creating the patterns.

The name itself is a hint: ‘Photo’ means light, and ‘resist’ means to withstand. In other words, it’s a material that ‘resists’ certain processes after being exposed to light. The entire process of drawing circuits on a wafer using this principle is called ‘photolithography’.

The 5 Steps of Photolithography at a Glance

  1. Coating: A thin, uniform layer of liquid photoresist is applied to the wafer. (Spin-coating)
  2. Exposure: A mask with the circuit blueprint is placed over the wafer, which is then exposed to ultraviolet (UV) light.
  3. Development: A developer solution selectively dissolves either the exposed or unexposed parts of the photoresist to create the pattern.
  4. Etching: The remaining photoresist acts as a protective barrier while the underlying layer is carved away.
  5. Stripping: Finally, the remaining photoresist is removed, leaving the finished circuit pattern.

Photoresists come in two types: ‘positive-tone,’ which dissolves when exposed to light, and ‘negative-tone,’ which hardens. While negative-tone was developed first, it had an issue where it would slightly swell during the hardening process, which reduced precision in ultra-fine circuits. In contrast, positive-tone resists don't have this problem, allowing for much more precise patterns. That’s why almost all modern processes use positive-tone photoresists.

 

From Asphalt to Advanced Materials: The History of Photoresist ๐Ÿ“œ

Amazingly, the origins of this technology date back to the 1820s in France. An inventor named Nicรฉphore Niรฉpce used a substance similar to asphalt, ‘Bitumen of Judea,’ which hardens when exposed to light, to create the world’s first photograph. This technique was adapted for semiconductors in the 1950s, thanks to a suggestion by William Shockley at Bell Labs, the inventor of the transistor.

Early photographic materials couldn’t withstand the harsh chemicals (like hydrofluoric acid) used in chip manufacturing. So, Kodak, famous for its camera film, stepped in to develop ‘KPR,’ a chemically resistant negative resist, and later ‘KTFR,’ which had better adhesion. KTFR became the industry standard for over 15 years.

๐Ÿ’ก TOK’s Game-Changing Move: The Story of OPR-800
But the real breakthrough was the arrival of positive-tone resist. The key was the ‘DNQ-Novolac’ system, originating from German blueprinting technology, which made much finer circuits possible. This tech spread to the US semiconductor industry, supposedly because a German company's American subsidiary was coincidentally located in the same New Jersey town as Bell Labs, leading to US and European firms dominating the market in the 1970s.

TOK moved incredibly fast. In 1979, it launched its decisive product: ‘OPR-800.’ While its performance was excellent, its true secret to success was its competitive price and the fact that it left less residue on wafers after use. This was a perfect match for the needs of Japan’s booming DRAM companies. They adopted OPR-800 en masse, allowing TOK to capture over 80% of the Japanese market. In a way, OPR-800 was the unsung hero behind Japan’s 1980s DRAM miracle.

 

The Rise of JSR and the Next Tech Leap ๐Ÿš€

While TOK dominated the market, another powerhouse quietly emerged: JSR. Originally a government-backed company making synthetic rubber for tires, JSR pivoted to electronic materials after the oil shock created a crisis. But JSR’s strategy was different.

JSR's Secret Sauce: Open Innovation and Global Strategy

The 1990s shift to DUV was a game-changer, and the solution was IBM’s ‘Chemically Amplified Resist (CAR).’ However, the technology was too sensitive for easy commercialization. Instead of hoarding it, IBM chose ‘open innovation,’ seeking collaboration with firms like JSR and TOK.

For JSR, this was a golden opportunity. At a time when Japan's semiconductor industry was slowing down, JSR used the partnership with IBM as a springboard to get ahead in the race to commercialize ArF photoresist. Building on this collaboration, JSR expanded its portfolio all the way to EUV and secured top-tier overseas clients like Samsung and Intel. By the early 2000s, it had become a true global player, with 70% of its revenue coming from international sales. Boldly leveraging technological partnership to conquer the global market was the core of JSR’s success story.

 

Photoresist at the Center of Geopolitics: The 2019 Japan-Korea Trade Dispute ๐ŸŒ

The strategic importance of photoresist was thrust onto the world stage during the July 2019 trade dispute between Japan and South Korea. Let’s take a closer, neutral look at what happened.

The Dispute and the Stated Positions

  • Japan's Action: In July 2019, the Japanese government tightened export procedures for three materials to South Korea: EUV photoresist, high-purity hydrogen fluoride (HF), and fluorinated polyimides (PI).
  • Japan's Official Stance: The stated reason was national security, citing concerns that these strategic materials could be diverted for military use and that South Korea's export control systems were inadequate.
  • South Korea's Official Stance: It strongly protested the move, framing it as ‘economic retaliation’ for a 2018 South Korean Supreme Court ruling regarding compensation for wartime forced laborers.

So, what was the actual impact on the semiconductor industry? Ultimately, the feared worst-case scenario of a ‘production line shutdown’ never happened.

๐Ÿ’ก Why the Impact Was Limited
The regulations were narrowly focused on EUV photoresist, a cutting-edge technology at the time. Both Samsung and SK Hynix were still in the early stages of adopting EUV, so they didn't require large volumes for mass production immediately. Furthermore, suppliers like JSR had alternative supply routes through joint ventures, such as with IMEC in Belgium, preventing a complete supply chain collapse. The restrictions were eventually lifted in 2023 as relations between the two countries improved.

 

How Does Japan Dominate the Market? ๐Ÿ‡ฏ๐Ÿ‡ต

The dispute is over, but a fundamental question remains: even in the EUV era, how does Japan maintain its absolute leadership in photoresist? The secret isn’t just one thing but a combination of five powerful factors.

Japan's 5 Keys to Photoresist Success
1. Manufacturing Clusters Key players like TOK and JSR are geographically concentrated, creating a hotbed of innovation through the exchange of talent, tech, and information. This environment of competition and cooperation accelerates development.
2. Open Innovation They masterfully adopted external technologies, like IBM's CAR, and evolved from being mere adopters to indispensable co-development partners, fully internalizing the tech.
3. Customer Co-Development Advanced photoresists are not off-the-shelf products. They are custom-tailored solutions developed jointly with clients like Samsung and TSMC, creating a powerful barrier to entry due to massive switching costs.
4. Long-Term Relationships A business culture that prioritizes long-term trust and sustainable partnerships over short-term profits has built incredible stability and customer loyalty.
5. Extreme Quality Control The purity required is astounding—akin to allowing only one drop of impurity in two Olympic-sized swimming pools. This level of quality, built on decades of know-how, is nearly impossible to replicate quickly.
Heads Up! The Market Paradox: Small but Critical
Despite its strategic importance, the photoresist market is tiny compared to the overall semiconductor industry (around $2 billion), and profit margins are not high (JSR ~3.8%, TOK ~7.8%). This creates a ‘high-risk, low-return’ structure, making it vulnerable to outside acquisition attempts. A former JSR chairman once joked that it was ‘smaller than the ramen market in Japan.’

 

A National Asset: The Meaning of the JSR Takeover ๐Ÿข

This structural vulnerability eventually became a reality. After a failed acquisition attempt of JSR by Germany’s Merck in 2022 and continued pressure from activist funds, the Japanese government made an unprecedented move in 2023.

A government-backed fund (JIC) invested approximately $6 billion to acquire JSR, a healthy, profitable private company, and take it private. This was not a bailout; it was a clear declaration that Japan considers photoresist technology a core national asset essential for economic security and sovereignty, and that it would shield it from foreign threats.

๐Ÿ’ก

The Photoresist Story: Key Takeaways

Photographic Origins: It evolved from 19th-century asphalt photography into a critical semiconductor material.
Japan's Winning Strategy: Dominance came not just from tech, but from clusters, open innovation, and deep customer integration.
An Industrial Paradox:
‘Low Profit + High Barrier’ created a vulnerability despite its strategic importance.
The Era of Tech Nationalism: The Japanese government's takeover of JSR proves this liquid is now a national strategic asset.

 

Frequently Asked Questions ❓

Q: How does talent exchange specifically work in Japan's photoresist cluster?

Talent exchange in Japan's photoresist industry is centered around building an ‘Open Innovation Ecosystem.’ Its key feature is the active use of global hubs rather than being confined to a specific region.

1. Global Collaboration at Albany NanoTech Complex: Scientists and engineers from Rapidus collaborate on next-gen technology with global firms like IBM, Samsung Electronics, JSR, and universities at the Albany NanoTech Complex in New York.

2. Rapidus-IBM Partnership: Rapidus is sending over 100 engineers to IBM's facilities to master Gate-All-Around (GAA) technology, crucial for the 2nm process, while also actively recruiting veteran semiconductor engineers within Japan.

3. Research Collaboration with IMEC in Belgium: They leverage international open innovation research hubs by collaborating with world-renowned semiconductor research center IMEC in Belgium.

Q: Are there successful B2B co-development cases in Korea's semiconductor materials sector?

Yes, the most prominent success story is the co-development of EUV photoresist between Dongjin Semichem and Samsung Electronics.

1. Domestic Success: They succeeded in developing EUV photoresist, one of the three items restricted by Japan in 2019, marking a major milestone in technological self-sufficiency.

2. Rapid Implementation: Samsung applied Dongjin Semichem's EUV PR to its mass production lines less than a year after it passed reliability tests, showcasing the success of their close collaboration.

3. Infrastructure and Global Collaboration: Dongjin Semichem made bold investments in its own lithography equipment and forged a partnership with IMEC in Belgium. Building on this, it has become the world's No. 1 supplier of PR for 3D NAND flash, with over 35% market share.

Q: How can Japan's open innovation model be adapted to the Korean context?

[Features of Japan's Open Innovation Model]

1. Consortium-Based Collaboration: ‘Rapidus,’ established in August 2022 with backing from eight major corporations including Toyota and Sony, aims to develop 2nm process technology by 2027, serving as a prime example of a national-level collaborative model.

2. Public-Private Partnership: Since 2021, the Japanese government has used large-scale subsidies to attract global giants like TSMC and Micron while also supporting domestic firms like Kioxia, rapidly restoring its domestic production base.

[Application for Korea]

1. National Strategic Approach: Korea must also recognize the semiconductor industry as a ‘survival strategy’ essential for economic security and move beyond short-term tax credits to establish a robust, long-term financial support system including subsidies, loans, and infrastructure.

2. Fostering Open Innovation: To truly succeed in domesticating materials and equipment, it's crucial to strengthen the quality of private-sector companies to a level that surpasses foreign leaders. This should be an opportunity to advance technological capabilities through open innovation, independent of external policies.

3. A Korean-Style CREATE Model: A six-point CREATE policy is proposed to make Korea a leader in open innovation where the creativity of startups and the global competitiveness of large corporations create synergy. This includes expanding funding for Proof-of-Concept (PoC) and matching funds, and increasing deal-sourcing opportunities.

And that’s a wrap! The story of how the liquid in a tiny bottle is shaping global geopolitics is pretty incredible, isn’t it? It will be fascinating to see how other small but strategically vital industries evolve in the future. If you have any more questions, feel free to ask in the comments! ๐Ÿ˜Š

๋ฐ˜๋„์ฒด ํŒจ๊ถŒ์˜ ์—ด์‡ , ํฌํ† ๋ ˆ์ง€์ŠคํŠธ: ์ผ๋ณธ์€ ์–ด๋–ป๊ฒŒ ์„ธ๊ณ„ 1์œ„๊ฐ€ ๋˜์—ˆ๋‚˜?

 

์ด ์ž‘์€ ๋ณ‘ ์† ์•ก์ฒด๊ฐ€ ์–ด๋–ป๊ฒŒ ๋ฐ˜๋„์ฒด ํŒจ๊ถŒ์˜ ์—ด์‡ ๊ฐ€ ๋์„๊นŒ์š”? ๋ฐ˜๋„์ฒด ๊ณต์ •์˜ ์ˆจ์€ ์ฃผ์—ญ, ‘ํฌํ† ๋ ˆ์ง€์ŠคํŠธ’. ์ผ๋ณธ ๊ธฐ์—…๋“ค์ด ์„ธ๊ณ„ ์‹œ์žฅ์˜ 90%๋ฅผ ์žฅ์•…ํ•˜๊ฒŒ ๋œ ๋น„๋ฐ€์„ ๊ทธ ์‹œ์ž‘๋ถ€ํ„ฐ ์ตœ์‹  ๊ธฐ์ˆ , ๊ทธ๋ฆฌ๊ณ  ์ง€์ •ํ•™์  ๋งฅ๋ฝ๊นŒ์ง€ ๊นŠ์ด ํŒŒํ—ค์ณ ๋ด…๋‹ˆ๋‹ค.

์•ˆ๋…•ํ•˜์„ธ์š”! ์˜ค๋Š˜์€ ๋ฐ˜๋„์ฒด ๋งŒ๋“œ๋Š” ๋ฐ ๊ผญ ํ•„์š”ํ•œ๋ฐ ์ด๋ฆ„์€ ์ข€ ๋‚ฏ์„ , ๊ทธ๋Ÿฐ ํ™”ํ•™ ๋ฌผ์งˆ์— ๋Œ€ํ•œ ์ด์•ผ๊ธฐ์˜ˆ์š”. ๋ฐ”๋กœ ‘ํฌํ† ๋ ˆ์ง€์ŠคํŠธ’์ธ๋ฐ์š”. ํ˜น์‹œ ๋“ค์–ด๋ณด์…จ๋‚˜์š”? ์ด๊ฒŒ ์ •๋ง ํฅ๋ฏธ๋กœ์šด ์ฃผ์ œ์ธ ๊ฒŒ, ์ผ๋ณธ ๊ธฐ์—…๋“ค์ด ์ด ์‹œ์žฅ์˜ ๊ฑฐ์˜ 90%๋ฅผ ์žฅ์•…ํ•˜๊ณ  ์žˆ๋‹ค๋Š” ์‚ฌ์‹ค์ด์—์š”. ๊ฑฐ์˜ ๋…์ ์ด์ฃ .

๊ทธ๋ž˜์„œ ์˜ค๋Š˜, ์ด ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ๊ธฐ์ˆ ์ด ์–ด๋–ป๊ฒŒ ๋ฐœ์ „ํ•ด ์™”๊ณ , ์ผ๋ณธ ๊ธฐ์—…๋“ค์ด ์–ด๋–ป๊ฒŒ ์ง€๊ธˆ์˜ ์œ„์น˜๊นŒ์ง€ ์˜ค๊ฒŒ ๋๋Š”์ง€, ๊ทธ ์—ญ์‚ฌ๋ถ€ํ„ฐ ์ „๋žต, ์ง€์ •ํ•™์  ๋งฅ๋ฝ๊นŒ์ง€ ๊นŠ์ด ํŒŒ๊ณ ๋“ค์–ด ํ•ต์‹ฌ์„ ๋ณด๋ ค๊ณ  ํ•ด์š”. ์ด ๋ณต์žกํ•œ ์ด์•ผ๊ธฐ๋ฅผ ๋น ๋ฅด๋ฉด์„œ๋„ ๊นŠ์ด ์žˆ๊ฒŒ ์ดํ•ดํ•˜์‹ค ์ˆ˜ ์žˆ๋„๋ก ๋„์™€๋“œ๋ฆด๊ฒŒ์š”! ๐Ÿ˜Š

 

ํฌํ† ๋ ˆ์ง€์ŠคํŠธ, ๋Œ€์ฒด ์ •์ฒด๊ฐ€ ๋ญ”๊ฐ€์š”? ๐Ÿค”

์‰ฝ๊ฒŒ ๋งํ•˜๋ฉด, ๋ฐ˜๋„์ฒด ์›จ์ดํผ ์œ„์— ์•„์ฃผ ๋ฏธ์„ธํ•œ ํšŒ๋กœ ํŒจํ„ด์„ ์ƒˆ๊ธธ ๋•Œ ์“ฐ๋Š” ‘๊ฐ๊ด‘์•ก’ ๊ฐ™์€ ๊ฑฐ์˜ˆ์š”. ์‚ฌ์ง„ ํ˜„์ƒํ•  ๋•Œ ์“ฐ๋Š” ํ•„๋ฆ„์ฒ˜๋Ÿผ ๋น›์— ๋ฐ˜์‘ํ•˜๋Š” ๊ฑฐ์ฃ . ์ด๊ฒŒ ์—†์œผ๋ฉด ์ˆ˜์ฒœ์–ต ์›์งœ๋ฆฌ ์ตœ์ฒจ๋‹จ ๋…ธ๊ด‘ ์žฅ๋น„๋„ ๊ทธ๋ƒฅ ๊ณ ์ฒ  ๋ฉ์–ด๋ฆฌ์— ๋ถˆ๊ณผํ•˜๋‹ค๊ณ  ํ•ด์š”. ๋ญ๋ž„๊นŒ, ์•„๋ฌด๋ฆฌ ์ข‹์€ ํ”„๋ฆฐํ„ฐ๊ฐ€ ์žˆ์–ด๋„ ์ž‰ํฌ๊ฐ€ ์—†์œผ๋ฉด ๋ชป ์“ฐ๋Š” ๊ฑฐ๋ž‘ ๋˜‘๊ฐ™์•„์š”. ์ด ํฌํ† ๋ ˆ์ง€์ŠคํŠธ๊ฐ€ ๋ฐ”๋กœ ๊ทธ ์ž‰ํฌ๋‚˜ ํŒจํ„ด์„ ์ฐ๋Š” ํ‹€ ์—ญํ• ์„ ํ•˜๋Š” ๊ฒ๋‹ˆ๋‹ค.

์ด๋ฆ„๋ถ€ํ„ฐ๊ฐ€ ํžŒํŠธ์ธ๋ฐ์š”, ‘ํฌํ† (Photo)’๋Š” ๋น›, ‘๋ ˆ์ง€์ŠคํŠธ(Resist)’๋Š” ์ €ํ•ญํ•œ๋‹ค๋Š” ๋œป์ด์—์š”. ์ฆ‰, ‘๋น›์— ๋ฐ˜์‘ํ•ด์„œ (ํŠน์ • ๊ณต์ •์—) ์ €ํ•ญํ•˜๋Š” ๋ฌผ์งˆ’์ด๋ผ๋Š” ์˜๋ฏธ์ฃ . ์ด ์›๋ฆฌ๋ฅผ ์ด์šฉํ•ด ์›จ์ดํผ ์œ„์— ํšŒ๋กœ๋ฅผ ๊ทธ๋ฆฌ๋Š” ๊ณผ์ •์„ ‘ํฌํ† ๊ณต์ •’ ๋˜๋Š” ‘๋ฆฌ์†Œ๊ทธ๋ž˜ํ”ผ’๋ผ๊ณ  ๋ถ€๋ฆ…๋‹ˆ๋‹ค.

๊ฐ„๋‹จํžˆ ๋ณด๋Š” ํฌํ†  ๊ณต์ • 5๋‹จ๊ณ„

  1. ๋„ํฌ(Coating): ์•ก์ฒด ์ƒํƒœ์˜ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ๋ฅผ ์›จ์ดํผ ์œ„์— ์–‡๊ณ  ๊ท ์ผํ•˜๊ฒŒ ๋ฐœ๋ผ์ค๋‹ˆ๋‹ค. (์Šคํ•€ ์ฝ”ํŒ…)
  2. ๋…ธ๊ด‘(Exposure): ํšŒ๋กœ๋„๊ฐ€ ๊ทธ๋ ค์ง„ ๋งˆ์Šคํฌ๋ฅผ ๋Œ€๊ณ  ์ž์™ธ์„ (UV) ๋น›์„ ์ชผ์—ฌ์ค๋‹ˆ๋‹ค.
  3. ํ˜„์ƒ(Development): ํ˜„์ƒ์•ก์œผ๋กœ ๋น›์„ ๋ฐ›์€ ๋ถ€๋ถ„ ๋˜๋Š” ๋ฐ›์ง€ ์•Š์€ ๋ถ€๋ถ„์„ ์„ ํƒ์ ์œผ๋กœ ๋…น์—ฌ๋‚ด ํŒจํ„ด์„ ๋งŒ๋“ญ๋‹ˆ๋‹ค.
  4. ์‹๊ฐ(Etching): ๋‚จ์€ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ๋ฅผ ๋ณดํ˜ธ๋ง‰ ์‚ผ์•„ ์•„๋ž˜์ธต ๋ง‰์„ ๊นŽ์•„๋ƒ…๋‹ˆ๋‹ค.
  5. ๋ฐ•๋ฆฌ(Stripping): ๋งˆ์ง€๋ง‰์œผ๋กœ ์ž„๋ฌด๋ฅผ ๋‹คํ•œ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ๋ฅผ ์ œ๊ฑฐํ•˜๋ฉด ํšŒ๋กœ ํŒจํ„ด์ด ์™„์„ฑ๋ฉ๋‹ˆ๋‹ค.

ํฌํ† ๋ ˆ์ง€์ŠคํŠธ๋Š” ๋น›์„ ๋ฐ›์•˜์„ ๋•Œ ๋…น๋Š” ‘ํฌ์ง€ํ‹ฐ๋ธŒ(Positive) ํƒ€์ž…’๊ณผ, ๋ฐ˜๋Œ€๋กœ ๋‹จ๋‹จํ•˜๊ฒŒ ๊ตณ๋Š” ‘๋„ค๊ฑฐํ‹ฐ๋ธŒ(Negative) ํƒ€์ž…’์œผ๋กœ ๋‚˜๋‰˜์–ด์š”. ์ดˆ๊ธฐ์—๋Š” ๋„ค๊ฑฐํ‹ฐ๋ธŒ ํƒ€์ž…์ด ๋จผ์ € ๊ฐœ๋ฐœ๋˜์—ˆ์ง€๋งŒ, ๋น›์„ ๋ฐ›์•„ ๊ตณ๋Š” ๊ณผ์ •์—์„œ ๋ฏธ์„ธํ•˜๊ฒŒ ๋ถ€ํ’€์–ด ์˜ค๋ฅด๋Š” ํ˜„์ƒ์ด ์žˆ์—ˆ์–ด์š”. ์ด๊ฒŒ ์ดˆ๋ฏธ์„ธ ๊ณต์ •์—์„œ๋Š” ์ •๋ฐ€๋„๋ฅผ ๋–จ์–ด๋œจ๋ฆฌ๋Š” ์›์ธ์ด ๋์ฃ . ๋ฐ˜๋ฉด์— ํฌ์ง€ํ‹ฐ๋ธŒ ํƒ€์ž…์€ ๊ทธ๋Ÿฐ ๋ฌธ์ œ๊ฐ€ ์ ์–ด ํ›จ์”ฌ ๋” ์ •๋ฐ€ํ•œ ํŒจํ„ด์„ ๋งŒ๋“ค ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์— ์š”์ฆ˜์€ ๊ฑฐ์˜ ๋‹ค ํฌ์ง€ํ‹ฐ๋ธŒ ํƒ€์ž…์„ ์‚ฌ์šฉํ•œ๋‹ต๋‹ˆ๋‹ค.

 

์•„์ŠคํŒ”ํŠธ์—์„œ ์ตœ์ฒจ๋‹จ ์†Œ์žฌ๊นŒ์ง€: ํฌํ† ๋ ˆ์ง€์ŠคํŠธ์˜ ์—ญ์‚ฌ ๐Ÿ“œ

์ •๋ง ๋†€๋ž๊ฒŒ๋„, ์ด ๊ธฐ์ˆ ์˜ ์‹œ์ž‘์€ 1820๋…„๋Œ€ ํ”„๋ž‘์Šค์—์„œ ์‹œ์ž‘๋์–ด์š”. ๋ฐœ๋ช…๊ฐ€ ๋‹ˆ์„ธํฌ๋ฅด ๋‹ˆ์—ก์Šค๊ฐ€ ์•„์ŠคํŒ”ํŠธ์™€ ๋น„์Šทํ•œ ‘์œ ๋Œ€ ๋น„ํˆฌ๋งจ’์ด๋ผ๋Š” ๋ฌผ์งˆ์ด ๋น›์„ ๋ฐ›์œผ๋ฉด ๊ตณ๋Š” ์„ฑ์งˆ์„ ์ด์šฉํ•ด ์„ธ๊ณ„ ์ตœ์ดˆ์˜ ์‚ฌ์ง„์„ ๋งŒ๋“  ๊ฒƒ์ด ์‹œ์ดˆ์˜€์ฃ . ์ด ๊ธฐ์ˆ ์ด ๋ฐ˜๋„์ฒด์— ์ ์šฉ๋œ ๊ฑด 1950๋…„๋Œ€, ํŠธ๋žœ์ง€์Šคํ„ฐ๋ฅผ ๋ฐœ๋ช…ํ•œ ๋ฒจ ์—ฐ๊ตฌ์†Œ์˜ ์œŒ๋ฆฌ์—„ ์‡ผํด๋ฆฌ์˜ ์ œ์•ˆ ๋•๋ถ„์ด์—ˆ์Šต๋‹ˆ๋‹ค.

์ฒ˜์Œ์—๋Š” ์‚ฌ์ง„์šฉ ๊ฐ๊ด‘์žฌ๋ฅผ ์ผ์ง€๋งŒ, ๋ฐ˜๋„์ฒด ๊ณต์ •์˜ ๋…ํ•œ ํ™”ํ•™๋ฌผ์งˆ(๋ถˆ์‚ฐ ๋“ฑ)์„ ๊ฒฌ๋””์ง€ ๋ชปํ–ˆ์–ด์š”. ๊ทธ๋ž˜์„œ ์นด๋ฉ”๋ผ ํ•„๋ฆ„์œผ๋กœ ์œ ๋ช…ํ•œ ์ฝ”๋‹ฅ(Kodak)์ด ๋‚˜์„œ์„œ ๋‚ดํ™”ํ•™์„ฑ์ด ๊ฐ•ํ•œ ๋„ค๊ฑฐํ‹ฐ๋ธŒ ๋ ˆ์ง€์ŠคํŠธ ‘KPR’๊ณผ ์ ‘์ฐฉ๋ ฅ์„ ๋†’์ธ ‘KTFR’์„ ๊ฐœ๋ฐœํ–ˆ๊ณ , ์ด KTFR์ด 15๋…„ ๋„˜๊ฒŒ ์—…๊ณ„ ํ‘œ์ค€์œผ๋กœ ์ž๋ฆฌ ์žก์•˜์ฃ .

๐Ÿ’ก TOK์˜ ๊ฒฐ์ •์  ํ•œ ๋ฐฉ: OPR-800์˜ ์„ฑ๊ณต ๋น„ํ™”
ํ•˜์ง€๋งŒ ์ง„์งœ ํŒ๋„๋ฅผ ๋ฐ”๊พผ ๊ฑด ํฌ์ง€ํ‹ฐ๋ธŒ ๋ ˆ์ง€์ŠคํŠธ์˜ ๋“ฑ์žฅ์ด์—ˆ์Šต๋‹ˆ๋‹ค. ๋…์ผ์˜ ์ฒญ์‚ฌ์ง„ ๊ธฐ์ˆ ์—์„œ ์œ ๋ž˜ํ•œ ‘DNQ-Novolac’ ์‹œ์Šคํ…œ ๋•๋ถ„์— ํ›จ์”ฌ ์ •๋ฐ€ํ•œ ํšŒ๋กœ๋ฅผ ๊ทธ๋ฆด ์ˆ˜ ์žˆ๊ฒŒ ๋œ ๊ฒƒ์ด์ฃ . ์ด ๊ธฐ์ˆ ์€ ๋ฒจ ์—ฐ๊ตฌ์†Œ์™€ ๊ฐ™์€ ๋™๋„ค์— ์žˆ๋˜ ๋…์ผ ํšŒ์‚ฌ์˜ ๋ฏธ๊ตญ ์žํšŒ์‚ฌ๋ฅผ ํ†ตํ•ด ๋ฏธ๊ตญ ๋ฐ˜๋„์ฒด ์—…๊ณ„์— ์ „ํŒŒ๋˜์—ˆ๊ณ , 1970๋…„๋Œ€์—๋Š” ๋ฏธ๊ตญ/์œ ๋Ÿฝ ๊ธฐ์—…๋“ค์ด ์‹œ์žฅ์„ ์ฃผ๋„ํ–ˆ์Šต๋‹ˆ๋‹ค.

์ด๋•Œ TOK๋Š” ๋งค์šฐ ๋ฐœ ๋น ๋ฅด๊ฒŒ ์›€์ง์˜€์Šต๋‹ˆ๋‹ค. 1979๋…„, ๊ฒฐ์ •์ ์ธ ์ œํ’ˆ์ธ ‘OPR-800’์„ ์ถœ์‹œํ•ฉ๋‹ˆ๋‹ค. ์„ฑ๋Šฅ๋„ ๋›ฐ์–ด๋‚ฌ์ง€๋งŒ, ์ง„์งœ ์„ฑ๊ณต ๋น„๊ฒฐ์€ ๊ฐ€๊ฒฉ ๊ฒฝ์Ÿ๋ ฅ๊ณผ ์‚ฌ์šฉ ํ›„ ์›จ์ดํผ์— ์ž”๋ฅ˜๋ฌผ์ด ์ ๋‹ค๋Š” ์žฅ์ ์ด์—ˆ์Šต๋‹ˆ๋‹ค. ์ด๋Š” ๋‹น์‹œ ํญ๋ฐœ์ ์œผ๋กœ ์„ฑ์žฅํ•˜๋˜ ์ผ๋ณธ DRAM ํšŒ์‚ฌ๋“ค์˜ ์š”๊ตฌ์— ์™„๋ฒฝํ•˜๊ฒŒ ๋ถ€ํ•ฉํ–ˆ๊ณ , ๊ทธ ๊ฒฐ๊ณผ TOK๋Š” ์ผ๋ณธ ์‹œ์žฅ์˜ 80% ์ด์ƒ์„ ์žฅ์•…ํ•˜๋ฉฐ ๊ฑฐ์ธ์œผ๋กœ ์„ฑ์žฅํ•ฉ๋‹ˆ๋‹ค. 1980๋…„๋Œ€ ์ผ๋ณธ DRAM ์‹ ํ™”์˜ ๋ฐ‘๋ฐ”ํƒ•์—๋Š” ๋ฐ”๋กœ ์ด OPR-800์ด ์žˆ์—ˆ๋˜ ์…ˆ์ด์ฃ .

 

๋˜ ๋‹ค๋ฅธ ๊ฐ•์ž, JSR์˜ ๋“ฑ์žฅ๊ณผ ๊ธฐ์ˆ  ํ˜์‹  ๐Ÿš€

TOK๊ฐ€ ์‹œ์žฅ์„ ์ฃผ๋„ํ•˜๋˜ ๋•Œ, ๋˜ ๋‹ค๋ฅธ ๊ฐ•์ž๊ฐ€ ์กฐ์šฉํžˆ ๋“ฑ์žฅํ•ฉ๋‹ˆ๋‹ค. ๋ฐ”๋กœ ‘JSR’์ด์—์š”. ์›๋ž˜ ํƒ€์ด์–ด์šฉ ํ•ฉ์„ฑ๊ณ ๋ฌด๋ฅผ ๋งŒ๋“ค๋˜ ์ •๋ถ€ ์ฃผ๋„ ๊ธฐ์—…์ด์—ˆ๋Š”๋ฐ, ์˜ค์ผ ์‡ผํฌ๋กœ ์œ„๊ธฐ๋ฅผ ๋งž์ž ์ „์ž์žฌ๋ฃŒ๋กœ ๋ˆˆ์„ ๋Œ๋ฆฐ ๊ฑฐ์ฃ . JSR์˜ ์„ฑ๊ณต ์ „๋žต์€ ๋‹ฌ๋ž์Šต๋‹ˆ๋‹ค.

JSR์˜ ์„ฑ๊ณต ๋น„๊ฒฐ: ์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜๊ณผ ๊ธ€๋กœ๋ฒŒ ์ „๋žต

1990๋…„๋Œ€, DUV ์‹œ๋Œ€๋กœ์˜ ์ „ํ™˜๊ณผ ํ•จ๊ป˜ IBM์ด ๊ฐœ๋ฐœํ•œ ‘ํ™”ํ•™ ์ฆํญํ˜• ๋ ˆ์ง€์ŠคํŠธ(CAR)’๋Š” ๊ฒŒ์ž„ ์ฒด์ธ์ €์˜€์Šต๋‹ˆ๋‹ค. ํ•˜์ง€๋งŒ ์ด ๊ธฐ์ˆ ์€ ๋„ˆ๋ฌด ์˜ˆ๋ฏผํ•ด์„œ ์ƒ์šฉํ™”๊ฐ€ ์–ด๋ ค์› ๊ณ , IBM์€ ๊ธฐ์ˆ ์„ ๋…์ ํ•˜๋Š” ๋Œ€์‹  ‘์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜’์„ ์„ ํƒ, JSR, TOK ๋“ฑ๊ณผ ํ˜‘๋ ฅ์„ ๋ชจ์ƒ‰ํ•ฉ๋‹ˆ๋‹ค.

JSR์—๊ฒŒ ์ด๊ฒƒ์€ ๊ฒฐ์ •์  ๊ธฐํšŒ์˜€์Šต๋‹ˆ๋‹ค. ์ผ๋ณธ ๋ฐ˜๋„์ฒด ์‚ฐ์—…์ด ์ฃผ์ถคํ•˜๋˜ ์‹œ๊ธฐ, JSR์€ IBM๊ณผ์˜ ๊ธฐ์ˆ  ํ˜‘๋ ฅ์„ ํ†ตํ•ด ArF ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ์ƒ์šฉํ™” ๊ฒฝ์Ÿ์—์„œ ์•ž์„œ ๋‚˜๊ฐˆ ๋ฐœํŒ์„ ๋งˆ๋ จํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด ํ˜‘๋ ฅ์„ ๊ธฐ๋ฐ˜์œผ๋กœ JSR์€ ํฌํŠธํด๋ฆฌ์˜ค๋ฅผ EUV๊นŒ์ง€ ํ™•์žฅํ–ˆ๊ณ , ์‚ผ์„ฑ์ „์ž, ์ธํ…” ๋“ฑ ํ•ด์™ธ ์„ ๋„ ๊ธฐ์—…๋“ค์„ ๊ณ ๊ฐ์œผ๋กœ ํ™•๋ณดํ•˜๋ฉฐ ๊ธ‰์„ฑ์žฅํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ทธ ๊ฒฐ๊ณผ 2000๋…„๋Œ€ ์ดˆ๋ฐ˜์—๋Š” ํ•ด์™ธ ๋งค์ถœ ๋น„์ค‘์ด 70%์— ์ด๋ฅผ ์ •๋„๋กœ ์™„๋ฒฝํ•œ ๊ธ€๋กœ๋ฒŒ ํ”Œ๋ ˆ์ด์–ด๋กœ ์ž๋ฆฌ ์žก์•˜์Šต๋‹ˆ๋‹ค. ๊ธฐ์ˆ  ํ˜‘๋ ฅ์„ ๋ฐœํŒ ์‚ผ์•„ ๊ณผ๊ฐํ•˜๊ฒŒ ๊ธ€๋กœ๋ฒŒ ์‹œ์žฅ์„ ๊ณต๋žตํ•œ ๊ฒƒ์ด JSR ์„ฑ๊ณต ์Šคํ† ๋ฆฌ์˜ ํ•ต์‹ฌ์ด์—ˆ์ฃ .

 

์ง€์ •ํ•™์˜ ์ค‘์‹ฌ์— ์„  ํฌํ† ๋ ˆ์ง€์ŠคํŠธ: 2019๋…„ ํ•œ์ผ ๋ฌด์—ญ ๋ถ„์Ÿ ๐ŸŒ

ํฌํ† ๋ ˆ์ง€์ŠคํŠธ์˜ ์ „๋žต์  ์ค‘์š”์„ฑ์ด ์ „ ์„ธ๊ณ„์— ๊ฐ์ธ๋œ ์‚ฌ๊ฑด์ด ์žˆ์—ˆ์ฃ . ๋ฐ”๋กœ 2019๋…„ 7์›”์— ์‹œ์ž‘๋œ ํ•œ์ผ ๋ฌด์—ญ ๋ถ„์Ÿ์ž…๋‹ˆ๋‹ค. ๋‹น์‹œ ์ƒํ™ฉ์„ ์ค‘๋ฆฝ์ ์ธ ์‹œ๊ฐ์—์„œ ๋‹ค์‹œ ํ•œ๋ฒˆ ์ž์„ธํžˆ ์งš์–ด๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค.

์‚ฌํƒœ์˜ ๋ฐœ๋‹จ๊ณผ ์–‘์ธก์˜ ์ž…์žฅ

  • ์ผ๋ณธ์˜ ์กฐ์น˜: 2019๋…„ 7์›”, ์ผ๋ณธ ์ •๋ถ€๋Š” EUV์šฉ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ, ๊ณ ์ˆœ๋„ ๋ถˆํ™”์ˆ˜์†Œ(HF), ํ”Œ๋ฃจ์˜ค๋ฆฐ ํด๋ฆฌ์ด๋ฏธ๋“œ(PI) 3๊ฐœ ํ’ˆ๋ชฉ์— ๋Œ€ํ•ด ํ•œ๊ตญ์œผ๋กœ์˜ ์ˆ˜์ถœ ์ ˆ์ฐจ๋ฅผ ๊ฐ•ํ™”ํ–ˆ์Šต๋‹ˆ๋‹ค.
  • ์ผ๋ณธ์˜ ๊ณต์‹ ์ž…์žฅ: ์ด๋Ÿฌํ•œ ์ „๋žต ๋ฌผ์ž๋“ค์ด ๊ตฐ์‚ฌ์  ๋ชฉ์ ์œผ๋กœ ์ „์šฉ๋  ๊ฐ€๋Šฅ์„ฑ์— ๋Œ€ํ•œ ์•ˆ๋ณด์ƒ์˜ ์šฐ๋ ค์™€, ํ•œ๊ตญ์˜ ์ˆ˜์ถœ ๊ด€๋ฆฌ ์‹œ์Šคํ…œ์ด ๋ฏธํกํ•˜๋‹ค๋Š” ์ ์„ ์ด์œ ๋กœ ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.
  • ํ•œ๊ตญ์˜ ๊ณต์‹ ์ž…์žฅ: 2018๋…„ ํ•œ๊ตญ ๋Œ€๋ฒ•์›์˜ ๊ฐ•์ œ์ง•์šฉ ํ”ผํ•ด์ž ๋ฐฐ์ƒ ํŒ๊ฒฐ์— ๋Œ€ํ•œ ์‚ฌ์‹ค์ƒ์˜ ‘๊ฒฝ์ œ ๋ณด๋ณต’ ์กฐ์น˜๋ผ๊ณ  ๊ทœ์ •ํ•˜๋ฉฐ ๊ฐ•ํ•˜๊ฒŒ ๋ฐ˜๋ฐœํ–ˆ์Šต๋‹ˆ๋‹ค.

๊ทธ๋ ‡๋‹ค๋ฉด ์ด ์กฐ์น˜๋Š” ์‹ค์ œ ๋ฐ˜๋„์ฒด ์‚ฐ์—…์— ์–ด๋А ์ •๋„์˜ ์˜ํ–ฅ์„ ๋ฏธ์ณค์„๊นŒ์š”? ๊ฒฐ๋ก ๋ถ€ํ„ฐ ๋งํ•˜๋ฉด, ์šฐ๋ คํ–ˆ๋˜ ์ตœ์•…์˜ ‘์ƒ์‚ฐ ๋ผ์ธ ์ค‘๋‹จ’ ์‚ฌํƒœ๋Š” ์ผ์–ด๋‚˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค.

๐Ÿ’ก ์‹ค์ œ ์˜ํ–ฅ์ด ์ œํ•œ์ ์ด์—ˆ๋˜ ์ด์œ 
๊ทœ์ œ ๋Œ€์ƒ์ด ๋‹น์‹œ ์ตœ์ฒจ๋‹จ ๊ธฐ์ˆ ์ด์—ˆ๋˜ EUV ๊ณต์ •์šฉ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ์— ํ•œ์ •๋˜์—ˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ๋‹น์‹œ ์‚ผ์„ฑ์ „์ž๋‚˜ SKํ•˜์ด๋‹‰์Šค ๋ชจ๋‘ EUV ๊ณต์ •์„ ๋ง‰ ๋„์ž…ํ•˜๊ฑฐ๋‚˜ ํ…Œ์ŠคํŠธํ•˜๋Š” ์ดˆ๊ธฐ ๋‹จ๊ณ„์—ฌ์„œ, ๋‹น์žฅ ๋Œ€๋Ÿ‰ ์ƒ์‚ฐ์— ์ฐจ์งˆ์ด ์ƒ๊ธธ ์ •๋„์˜ ๋ฌผ๋Ÿ‰์ด ํ•„์š”ํ•˜์ง€๋Š” ์•Š์•˜์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ, JSR๊ณผ ๊ฐ™์€ ๊ณต๊ธ‰์‚ฌ๋Š” ๋ฒจ๊ธฐ์— IMEC๊ณผ์˜ ํ•ฉ์ž‘ ๋ฒ•์ธ์„ ํ†ตํ•œ ์šฐํšŒ ๊ณต๊ธ‰ ๊ฒฝ๋กœ๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ์–ด ๊ธ€๋กœ๋ฒŒ ๊ณต๊ธ‰๋ง์ด ์™„์ „ํžˆ ๋งˆ๋น„๋˜์ง€๋Š” ์•Š์•˜์Šต๋‹ˆ๋‹ค. ์ด ์กฐ์น˜๋Š” ๊ฒฐ๊ตญ 2023๋…„ ํ•œ์ผ ๊ด€๊ณ„๊ฐ€ ๊ฐœ์„ ๋˜๋ฉด์„œ ํ•ด์ œ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.

 

์ผ๋ณธ์€ ์–ด๋–ป๊ฒŒ ์‹œ์žฅ์„ ์ง€๋ฐฐํ•˜๊ฒŒ ๋์„๊นŒ? ๐Ÿ‡ฏ๐Ÿ‡ต

๋ถ„์Ÿ์€ ์ผ๋‹จ๋ฝ๋˜์—ˆ์ง€๋งŒ ๊ทผ๋ณธ์ ์ธ ์งˆ๋ฌธ์€ ๋‚จ์Šต๋‹ˆ๋‹ค. EUV ์‹œ๋Œ€์—๋„ ์ผ๋ณธ์€ ์–ด๋–ป๊ฒŒ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ์‹œ์žฅ์˜ ์ ˆ๋Œ€ ๊ฐ•์ž๋กœ ๊ตฐ๋ฆผํ•˜๊ณ  ์žˆ์„๊นŒ์š”? ๊ทธ ๋น„๊ฒฐ์€ ๋‹จ์ˆœํžˆ ๊ธฐ์ˆ ๋ ฅ ํ•˜๋‚˜๋งŒ์œผ๋กœ ์„ค๋ช…ํ•  ์ˆ˜ ์—†๋Š”, ๋‹ค์„ฏ ๊ฐ€์ง€ ๋ณตํ•ฉ์ ์ธ ์š”์ธ์— ์žˆ์Šต๋‹ˆ๋‹ค.

์ผ๋ณธ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ์‚ฐ์—…์˜ 5๊ฐ€์ง€ ์„ฑ๊ณต ๋ฐฉ์ •์‹
1. ์ œ์กฐ ํด๋Ÿฌ์Šคํ„ฐ์˜ ํž˜ TOK, JSR ๋“ฑ ์ฃผ์š” ๊ธฐ์—…๋“ค์ด ๊ฐ€๋‚˜๊ฐ€์™€ํ˜„ ๋“ฑ ์ˆ˜๋„๊ถŒ์— ๋ฐ€์ง‘ํ•ด ์ธ์žฌ, ๊ธฐ์ˆ , ์ •๋ณด ๊ต๋ฅ˜๋ฅผ ํ†ตํ•ด ํ˜์‹ ์„ ๊ฐ€์†ํ™”ํ–ˆ์Šต๋‹ˆ๋‹ค. ๊ฒฝ์Ÿ๊ณผ ํ˜‘๋ ฅ์ด ๊ณต์กดํ•˜๋ฉฐ ‘์–ด๊นจ๋„ˆ๋จธ ๋ฐฐ์šฐ๊ธฐ’๊ฐ€ ๊ฐ€๋Šฅํ•œ ํ™˜๊ฒฝ์ž…๋‹ˆ๋‹ค.
2. ์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜ ์™ธ๋ถ€ ๊ธฐ์ˆ (IBM์˜ CAR)์„ ์ ๊ทน์ ์œผ๋กœ ๋ฐ›์•„๋“ค์—ฌ ๋‹จ์ˆœ ์ฑ„ํƒ์„ ๋„˜์–ด ๊ณต๋™ ๊ฐœ๋ฐœ ํŒŒํŠธ๋„ˆ๋กœ ๋ฐœ์ „ํ•˜๋ฉฐ ๊ธฐ์ˆ ์„ ๋‚ด์žฌํ™”ํ–ˆ์Šต๋‹ˆ๋‹ค.
3. ๊ณ ๊ฐ ๋ฐ€์ฐฉ ๊ณต๋™๊ฐœ๋ฐœ ์ตœ์ฒจ๋‹จ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ๋Š” ๊ธฐ์„ฑํ’ˆ์ด ์•„๋‹™๋‹ˆ๋‹ค. ์‚ผ์„ฑ, TSMC ๋“ฑ ๊ณ ๊ฐ์‚ฌ์˜ ํŠน์ • ๊ณต์ • ๋ผ์ธ์— ๋งž์ถฐ ํ•จ๊ป˜ ๊ฐœ๋ฐœํ•˜๋Š” ‘๋งž์ถคํ˜• ์†”๋ฃจ์…˜’์œผ๋กœ, ํ•œ๋ฒˆ ์ ์šฉ๋˜๋ฉด ๊ต์ฒด๊ฐ€ ๊ฑฐ์˜ ๋ถˆ๊ฐ€๋Šฅํ•œ ๊ฐ•๋ ฅํ•œ ์ง„์ž… ์žฅ๋ฒฝ์„ ๋งŒ๋“ญ๋‹ˆ๋‹ค.
4. ์žฅ๊ธฐ์  ๊ด€๊ณ„ ์ค‘์‹œ ๋ฌธํ™” ๋‹จ๊ธฐ ์ˆ˜์ต๋ณด๋‹ค ๊ณ ๊ฐ๊ณผ์˜ ์‹ ๋ขฐ์™€ ์ง€์† ๊ฐ€๋Šฅํ•œ ํ˜‘๋ ฅ์„ ์šฐ์„ ์‹œํ•˜๋Š” ๋ฌธํ™”๊ฐ€ ์žฅ๊ธฐ์ ์ธ ๊ฒฝ์Ÿ๋ ฅ๊ณผ ๊ณ ๊ฐ ์ถฉ์„ฑ๋„๋ฅผ ํ™•๋ณดํ•˜๋Š” ๋น„๊ฒฐ์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
5. ๊ทนํ•œ์˜ ํ’ˆ์งˆ ๊ด€๋ฆฌ ์˜ฌ๋ฆผํ”ฝ ๊ทœ๊ฒฉ ์ˆ˜์˜์žฅ 2๊ฐœ ๋ถ„๋Ÿ‰์˜ ๋ฌผ์— ๋ถˆ์ˆœ๋ฌผ ๋‹จ ํ•œ ๋ฐฉ์šธ๋„ ์šฉ๋‚ฉํ•˜์ง€ ์•Š๋Š” ์ˆ˜์ค€์˜ ์ˆœ๋„ ๊ด€๋ฆฌ๊ฐ€ ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ์ˆ˜์‹ญ ๋…„๊ฐ„ ์ถ•์ ๋œ ์žฅ์ธ๊ธ‰ ๋…ธํ•˜์šฐ๋Š” ๋‹จ๊ธฐ๊ฐ„์— ๋”ฐ๋ผ์žก๊ธฐ ๋ถˆ๊ฐ€๋Šฅํ•œ ์˜์—ญ์ž…๋‹ˆ๋‹ค.
์ฃผ์˜ํ•˜์„ธ์š”! ์‹œ์žฅ์˜ ์—ญ์„ค: ์ž‘์ง€๋งŒ ์น˜๋ช…์ ์ด๋‹ค
์ด๋ ‡๊ฒŒ ์ „๋žต์ ์œผ๋กœ ์ค‘์š”ํ•˜์ง€๋งŒ, ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ์‹œ์žฅ์€ ๋ฐ˜๋„์ฒด ์ „์ฒด์— ๋น„ํ•˜๋ฉด ๊ทœ๋ชจ๊ฐ€ ๋งค์šฐ ์ž‘๊ณ (์•ฝ 20์–ต ๋‹ฌ๋Ÿฌ), ์˜์—… ์ด์ต๋ฅ ๋„ ๋†’์ง€ ์•Š์•„์š”. (JSR ์•ฝ 3.8%, TOK ์•ฝ 7.8%) ์ด๋Š” ‘๊ณ ์œ„ํ—˜ ์ €์ˆ˜์ต’ ๊ตฌ์กฐ๋กœ, ์™ธ๋ถ€์˜ ์ธ์ˆ˜ํ•ฉ๋ณ‘ ์‹œ๋„์— ์ทจ์•ฝํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์˜๋ฏธ์ด๊ธฐ๋„ ํ•ฉ๋‹ˆ๋‹ค. JSR ์ „ ํšŒ์žฅ์ด ๋†๋‹ด ์‚ผ์•„ ‘์ผ๋ณธ ๋ผ๋ฉ˜ ์‹œ์žฅ๋ณด๋‹ค ์ž‘๋‹ค’๊ณ  ํ–ˆ์„ ์ •๋„๋‹ˆ๊นŒ์š”.

 

๊ตญ๊ฐ€ ์ž์‚ฐ์ด ๋œ ๊ธฐ์ˆ : JSR ์ธ์ˆ˜ ์‚ฌํƒœ์˜ ์˜๋ฏธ ๐Ÿข

์ด๋Ÿฐ ๊ตฌ์กฐ์  ์ทจ์•ฝ์„ฑ์€ ๊ฒฐ๊ตญ ํ˜„์‹ค์ด ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. 2022๋…„ ๋…์ผ ๋จธํฌ์˜ JSR ์ธ์ˆ˜ ์‹œ๋„๊ฐ€ ๋ฌด์‚ฐ๋œ ์ดํ›„์—๋„ ์‚ฌ๋ชจํŽ€๋“œ๋“ค์˜ ๊ฒฝ์˜๊ถŒ ์œ„ํ˜‘์ด ๊ณ„์†๋˜์ž, ์ผ๋ณธ ์ •๋ถ€๋Š” 2023๋…„ ์•„์ฃผ ์ด๋ก€์ ์ธ ๊ฒฐ์ •์„ ๋‚ด๋ฆฝ๋‹ˆ๋‹ค.

๋ฐ”๋กœ ์ผ๋ณธ ์ •์ฑ…ํŽ€๋“œ(JIC)๊ฐ€ ์•ฝ 9์กฐ ์›์„ ํˆฌ์ž…ํ•ด ์šฐ๋Ÿ‰ ๋ฏผ๊ฐ„ ๊ธฐ์—…์ด๋˜ JSR์„ ์ธ์ˆ˜ํ•˜๊ณ  ๋น„์ƒ์žฅ ํšŒ์‚ฌ๋กœ ์ „ํ™˜ํ•˜๊ธฐ๋กœ ํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋Š” ๋ถ€์‹ค ๊ธฐ์—… ๊ตฌ์ œ๊ฐ€ ์•„๋‹Œ, ๊ตญ๊ฐ€ ์•ˆ๋ณด์™€ ๊ฒฝ์ œ ์ฃผ๊ถŒ์— ์ง๊ฒฐ๋œ ํ•ต์‹ฌ ๊ธฐ์ˆ ์„ ์™ธ๋ถ€ ์œ„ํ˜‘์œผ๋กœ๋ถ€ํ„ฐ ๋ณดํ˜ธํ•˜๋ ค๋Š” ๊ฐ•๋ ฅํ•œ ์˜์ง€์˜ ํ‘œํ˜„์ด์—ˆ์Šต๋‹ˆ๋‹ค.

๐Ÿ’ก

ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ์ด์•ผ๊ธฐ ํ•ต์‹ฌ ์š”์•ฝ

์‹œ์ž‘์€ ์‚ฌ์ง„ ๊ธฐ์ˆ : 19์„ธ๊ธฐ ์•„์ŠคํŒ”ํŠธ ์‚ฌ์ง„์—์„œ ์‹œ์ž‘ํ•ด ๋ฐ˜๋„์ฒด ํ•ต์‹ฌ ์†Œ์žฌ๋กœ ๋ฐœ์ „ํ–ˆ์Šต๋‹ˆ๋‹ค.
์ผ๋ณธ์˜ ์ง€๋ฐฐ ์ „๋žต: ๋‹จ์ˆœ ๊ธฐ์ˆ ๋ ฅ์„ ๋„˜์–ด ํด๋Ÿฌ์Šคํ„ฐ, ์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜, ๊ณ ๊ฐ ๋ฐ€์ฐฉ ๊ฐœ๋ฐœ์ด ํ•ต์‹ฌ์ด์—ˆ์Šต๋‹ˆ๋‹ค.
์—ญ์„ค์  ์‚ฐ์—… ๊ตฌ์กฐ:
‘๋‚ฎ์€ ์ˆ˜์ต์„ฑ + ๋†’์€ ๊ธฐ์ˆ  ์žฅ๋ฒฝ’์ด ์™ธ๋ถ€ ์œ„ํ˜‘์— ์ทจ์•ฝํ•˜๊ฒŒ ๋งŒ๋“ค์—ˆ์Šต๋‹ˆ๋‹ค.
๊ธฐ์ˆ  ์•ˆ๋ณด์˜ ์‹œ๋Œ€: ์ผ๋ณธ ์ •๋ถ€์˜ JSR ์ธ์ˆ˜๋Š” ์ž‘์€ ๋ณ‘ ์† ์•ก์ฒด๊ฐ€ ๊ตญ๊ฐ€ ์ „๋žต ์ž์‚ฐ์ด ๋˜์—ˆ์Œ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค.

 

์ž์ฃผ ๋ฌป๋Š” ์งˆ๋ฌธ ❓

Q: ์ผ๋ณธ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ํด๋Ÿฌ์Šคํ„ฐ์˜ ๊ตฌ์ฒด์ ์ธ ์ธ์žฌ ๊ต๋ฅ˜๋Š” ์–ด๋–ป๊ฒŒ ์ด๋ฃจ์–ด์ง€๋‚˜์š”?

์ผ๋ณธ ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ์‚ฐ์—…์˜ ์ธ์žฌ ๊ต๋ฅ˜๋Š” ‘์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜ ์ƒํƒœ๊ณ„’ ๊ตฌ์ถ•์„ ์ค‘์‹ฌ์œผ๋กœ ์ด๋ฃจ์–ด์ง‘๋‹ˆ๋‹ค. ํŠน์ • ์ง€์—ญ์— ๋ฌถ์ด๊ธฐ๋ณด๋‹ค๋Š” ๊ธ€๋กœ๋ฒŒ ๊ฑฐ์ ์„ ์ ๊ทน์ ์œผ๋กœ ํ™œ์šฉํ•˜๋Š” ๊ฒƒ์ด ํŠน์ง•์ž…๋‹ˆ๋‹ค.

1. Albany NanoTech Complex ์ค‘์‹ฌ์˜ ๊ธ€๋กœ๋ฒŒ ํ˜‘๋ ฅ: Rapidus์˜ ๊ณผํ•™์ž์™€ ์—”์ง€๋‹ˆ์–ด๋“ค์€ ๋‰ด์š•์˜ Albany NanoTech Complex์—์„œ IBM, ์‚ผ์„ฑ์ „์ž, JSR ๋“ฑ ๊ธ€๋กœ๋ฒŒ ๊ธฐ์—… ๋ฐ ๋Œ€ํ•™๋“ค๊ณผ ํ•จ๊ป˜ ์ฐจ์„ธ๋Œ€ ๊ธฐ์ˆ ์„ ๊ณต๋™ ์—ฐ๊ตฌํ•ฉ๋‹ˆ๋‹ค.

2. Rapidus-IBM ํŒŒํŠธ๋„ˆ์‹ญ: Rapidus๋Š” 100๋ช… ์ด์ƒ์˜ ์—”์ง€๋‹ˆ์–ด๋ฅผ IBM ์‹œ์„ค์— ํŒŒ๊ฒฌํ•ด 2nm ๊ณต์ •์˜ ํ•ต์‹ฌ์ธ GAA ๊ธฐ์ˆ ์„ ์Šต๋“ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ์ผ๋ณธ ๋‚ด ๋ฒ ํ…Œ๋ž‘ ๋ฐ˜๋„์ฒด ์—”์ง€๋‹ˆ์–ด๋„ ์ ๊ทน์ ์œผ๋กœ ์˜์ž…ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

3. ๋ฒจ๊ธฐ์— IMEC๊ณผ์˜ ์—ฐ๊ตฌ ํ˜‘๋ ฅ: ๋ฒจ๊ธฐ์—์˜ ์„ธ๊ณ„์ ์ธ ๋ฐ˜๋„์ฒด ์—ฐ๊ตฌ ํ—ˆ๋ธŒ IMEC๊ณผ์˜ ํ˜‘๋ ฅ์„ ํ†ตํ•ด ๊ตญ์ œ์ ์ธ ์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜ ์—ฐ๊ตฌ ๊ฑฐ์ ์„ ํ™œ์šฉํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

Q: ํ•œ๊ตญ ๋ฐ˜๋„์ฒด ์†Œ์žฌ ๋ถ„์•ผ์—์„œ ์„ฑ๊ณต์ ์ธ B2B ๊ณต๋™๊ฐœ๋ฐœ ์‚ฌ๋ก€๊ฐ€ ์žˆ๋‚˜์š”?

๋„ค, ๊ฐ€์žฅ ๋Œ€ํ‘œ์ ์ธ ์„ฑ๊ณต ์‚ฌ๋ก€๋Š” ๋™์ง„์Ž„๋ฏธ์ผ๊ณผ ์‚ผ์„ฑ์ „์ž์˜ EUV ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ๊ณต๋™๊ฐœ๋ฐœ์ž…๋‹ˆ๋‹ค.

1. ๊ตญ์‚ฐํ™” ์„ฑ๊ณต: 2019๋…„ ์ผ๋ณธ ์ˆ˜์ถœ ๊ทœ์ œ 3๋Œ€ ํ’ˆ๋ชฉ ์ค‘ ํ•˜๋‚˜์˜€๋˜ EUV ํฌํ† ๋ ˆ์ง€์ŠคํŠธ ๊ฐœ๋ฐœ์— ์„ฑ๊ณตํ•˜๋ฉฐ ๊ธฐ์ˆ  ์ž๋ฆฝ์˜ ์ค‘์š”ํ•œ ์ด์ •ํ‘œ๋ฅผ ์„ธ์› ์Šต๋‹ˆ๋‹ค.

2. ๋น ๋ฅธ ์–‘์‚ฐ ์ ์šฉ: ์‚ผ์„ฑ์ „์ž๋Š” ์‹ ๋ขฐ์„ฑ ์‹œํ—˜ ํ†ต๊ณผ 1๋…„์ด ์ฑ„ ๋˜์ง€ ์•Š์€ ์‹œ์ ์— ๋™์ง„์Ž„๋ฏธ์ผ์˜ EUV PR์„ ์‹ค์ œ ์–‘์‚ฐ ๋ผ์ธ์— ์ ์šฉํ•˜๋ฉฐ ๊ธด๋ฐ€ํ•œ ํ˜‘๋ ฅ์˜ ์„ฑ๊ณผ๋ฅผ ๋ณด์—ฌ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค.

3. ์ธํ”„๋ผ์™€ ๊ธ€๋กœ๋ฒŒ ํ˜‘๋ ฅ: ๋™์ง„์Ž„๋ฏธ์ผ์€ ์ž์ฒด ๋…ธ๊ด‘ ์žฅ๋น„๋ฅผ ๊ตฌ์ถ•ํ•˜๊ณ  ๋ฒจ๊ธฐ์— IMEC๊ณผ๋„ ํ˜‘๋ ฅ ๊ด€๊ณ„๋ฅผ ๋งบ๋Š” ๋“ฑ ๊ธฐ์ˆ  ๊ฐœ๋ฐœ์„ ์œ„ํ•œ ๊ณผ๊ฐํ•œ ํˆฌ์ž๋ฅผ ์ง„ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. ์ด๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ 3D ๋‚ธ๋“œํ”Œ๋ž˜์‹œ์šฉ PR ์‹œ์žฅ์—์„œ๋Š” ์ ์œ ์œจ 35% ์ด์ƒ์œผ๋กœ ์„ธ๊ณ„ 1์œ„๋ฅผ ๋‹ฌ์„ฑํ–ˆ์Šต๋‹ˆ๋‹ค.

Q: ์ผ๋ณธ์˜ ์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜ ๋ชจ๋ธ์„ ํ•œ๊ตญ ์ƒํ™ฉ์— ๋งž๊ฒŒ ์ ์šฉํ•˜๋Š” ๋ฐฉ์•ˆ์€ ๋ฌด์—‡์ธ๊ฐ€์š”?

[์ผ๋ณธ์˜ ์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜ ๋ชจ๋ธ ํŠน์ง•]

1. ์ปจ์†Œ์‹œ์—„ ๊ธฐ๋ฐ˜ ํ˜‘๋ ฅ: ‘Rapidus’๋Š” 2022๋…„ 8์›” ๋„์š”ํƒ€, ์†Œ๋‹ˆ ๋“ฑ 8๊ฐœ ์ฃผ์š” ๊ธฐ์—…์˜ ์ง€์›์œผ๋กœ ์„ค๋ฆฝ๋˜์–ด 2027๋…„๊นŒ์ง€ 2nm ๊ณต์ • ๊ฐœ๋ฐœ์„ ๋ชฉํ‘œ๋กœ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๊ฐœ๋ณ„ ๊ธฐ์—…์„ ๋„˜์–ด์„  ๊ตญ๊ฐ€์  ํ˜‘๋ ฅ ๋ชจ๋ธ์˜ ๋Œ€ํ‘œ์  ์‚ฌ๋ก€์ž…๋‹ˆ๋‹ค.

2. ์ •๋ถ€-๋ฏผ๊ฐ„ ํŒŒํŠธ๋„ˆ์‹ญ: ์ผ๋ณธ ์ •๋ถ€๋Š” 2021๋…„๋ถ€ํ„ฐ ๋Œ€๊ทœ๋ชจ ๋ณด์กฐ๊ธˆ์„ ํˆฌ์ž…ํ•ด TSMC, ๋งˆ์ดํฌ๋ก  ๋“ฑ ๊ธ€๋กœ๋ฒŒ ๊ธฐ์—…์„ ์œ ์น˜ํ•˜๊ณ , ํ‚ค์˜ฅ์‹œ์•„ ๋“ฑ ์ž๊ตญ ๊ธฐ์—…๋„ ์ง€์›ํ•˜๋ฉฐ ์ƒ์‚ฐ ๊ธฐ๋ฐ˜์„ ๋น ๋ฅด๊ฒŒ ๋ณต์›ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

[ํ•œ๊ตญ ์ ์šฉ ๋ฐฉ์•ˆ]

1. ๊ตญ๊ฐ€ ์ฐจ์›์˜ ์ „๋žต์  ์ ‘๊ทผ: ๋ฐ˜๋„์ฒด ์‚ฐ์—…์„ ‘์ƒ์กด ์ „๋žต’์œผ๋กœ ์ธ์‹ํ•˜๊ณ , ํˆฌ์ž ์„ธ์•ก๊ณต์ œ ์ค‘์‹ฌ์˜ ๋‹จ๊ธฐ ํ˜œํƒ์„ ๋„˜์–ด ๋ณด์กฐ๊ธˆ·๋Œ€์ถœ·์ธํ”„๋ผ ๋“ฑ ์‹คํšจ์„ฑ ์žˆ๋Š” ์ค‘์žฅ๊ธฐ ์žฌ์ • ์ง€์› ์ฒด๊ณ„๋ฅผ ๋งˆ๋ จํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.

2. ๊ฐœ๋ฐฉํ˜• ํ˜์‹  ํ™œ์„ฑํ™”: ์†Œ์žฌ·๋ถ€ํ’ˆ·์žฅ๋น„(์†Œ๋ถ€์žฅ) ๊ตญ์‚ฐํ™”๋ฅผ ์œ„ํ•ด ์™ธ๊ตญ ์„ ๋„๊ธฐ์—…์„ ์••๋„ํ•  ๋ฏผ๊ฐ„๊ธฐ์—… ์ฐจ์›์˜ ํ’ˆ์งˆ ๊ฐ•ํ™”๊ฐ€ ์ ˆ์‹คํ•˜๋ฉฐ, ๊ฐœ๋ฐฉํ˜• ํ˜์‹ ์„ ํ†ตํ•ด ์™ธ๋ถ€ ์ •์ฑ…์— ํ”๋“ค๋ฆฌ์ง€ ์•Š๋Š” ๊ธฐ์ˆ ๋ ฅ์„ ํ™•๋ณดํ•˜๋Š” ๊ณ„๊ธฐ๋กœ ์‚ผ์•„์•ผ ํ•ฉ๋‹ˆ๋‹ค.

3. ํ•œ๊ตญํ˜• CREATE ๋ชจ๋ธ: ์Šคํƒ€ํŠธ์—…์˜ ํ˜์‹ ๊ณผ ๋Œ€๊ธฐ์—…์˜ ๊ฒฝ์Ÿ๋ ฅ์ด ์‹œ๋„ˆ์ง€๋ฅผ ๋‚ด๋Š” ์˜คํ”ˆ ์ด๋…ธ๋ฒ ์ด์…˜ ์„ ๋„๊ตญ์œผ๋กœ ๋„์•ฝํ•˜๊ธฐ ์œ„ํ•ด PoC(๊ธฐ์ˆ ์‹ค์ฆ) ์ž๊ธˆ์ง€์› ๋ฐ ๋งค์นญํŽ€๋“œ ํ™•๋Œ€, ๋”œ ์†Œ์‹ฑ ๊ธฐํšŒ ํ™•๋Œ€ ๋“ฑ์˜ ์ •์ฑ…์ด ์ œ์•ˆ๋ฉ๋‹ˆ๋‹ค.

์˜ค๋Š˜ ์ด์•ผ๊ธฐ๋Š” ์—ฌ๊ธฐ๊นŒ์ง€์ž…๋‹ˆ๋‹ค. ์ž‘์€ ๋ณ‘ ์† ์•ก์ฒด๊ฐ€ ๋งŒ๋“ค์–ด๋‚ธ ๊ธ€๋กœ๋ฒŒ ์ง€์ •ํ•™, ์ •๋ง ํฅ๋ฏธ๋กญ์ง€ ์•Š๋‚˜์š”? ์•ž์œผ๋กœ JSR ์‚ฌ๋ก€์ฒ˜๋Ÿผ ์‹œ์žฅ ๊ทœ๋ชจ๋Š” ์ž‘์ง€๋งŒ ์ „๋žต์  ๊ฐ€์น˜๊ฐ€ ํฐ ๋ถ„์•ผ๋“ค์ด ์–ด๋–ป๊ฒŒ ๋ณ€ํ•ด๊ฐˆ์ง€ ์ง€์ผœ๋ณด๋Š” ๊ฒƒ๋„ ์žฌ๋ฏธ์žˆ๋Š” ๊ด€์ „ ํฌ์ธํŠธ๊ฐ€ ๋  ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค. ๋” ๊ถ๊ธˆํ•œ ์ ์ด ์žˆ๋‹ค๋ฉด ๋Œ“๊ธ€๋กœ ๋ฌผ์–ด๋ด์ฃผ์„ธ์š”~ ๐Ÿ˜Š

The Key to HBM Performance: How Hybrid Bonding Will Change Semiconductors

 

With HBM4 and the AI era upon us, why is everyone suddenly talking about ‘hybrid bonding?’

We’ll break down everything you need to know about this revolutionary packaging technology that gets rid of solder balls—from its core principles to the fierce nanometer-scale challenges, its difficult path to HBM integration, and what it means for the future.

Recently, the AI semiconductor market heated up once again with SK Hynix's announcement that they’ve successfully developed and started mass production of HBM4. The news had experts and investors focused on a single question: ‘Did they actually use the so-called “dream technology,” hybrid bonding, in this version of HBM4?’

The short answer is, not yet. It appears that the initial production of HBM4 will use an advanced version of existing technology (MR-MUF), while hybrid bonding is still being developed as a ‘key future technology’ for ultra-high-stack HBM with 16 or more layers, or for the next generation of memory. However, hybrid bonding has moved beyond being just an option; it's now a critical turning point in the semiconductor packaging race.

My background is in mechanical engineering, but I’ve also studied law and worked on a master's in AI computing, handling numerous patents in the memory semiconductor industry. Through this, I’ve come to a firm belief: ‘The more complex the technology, the more crucial it is to explain it in a way that more people can understand.’ This article is my attempt to build a small bridge between the technology and the market.

 

1. Why the Sudden Focus on Advanced Packaging?

The game of semiconductor performance is changing. The competition is no longer just about how finely you can etch circuits inside a chip. The focus is shifting to ‘how well you can connect and stack’ those chips—in other words, packaging.

The biggest reason for this shift is that ‘Moore's Law’ isn't what it used to be. The cost and technical difficulty of making circuits smaller have skyrocketed. So, it's now more efficient, both in terms of performance and cost, to create smaller, specialized chips called ‘chiplets’ and then assemble them like LEGOs.

Especially in fields like AI and High-Performance Computing (HPC), which need to process staggering amounts of data, how quickly and efficiently you can connect these chiplets has become the key factor that determines performance.

๐Ÿ’ก So, what was wrong with the old way?
The traditional method using ‘solder bumps’ has clear physical limitations. The spacing (pitch) of these tiny solder balls is measured in tens of micrometers, and their size makes it incredibly difficult to dramatically increase the number of data pathways (I/O density). Technologies like SK Hynix’s MR-MUF are improvements, but they are still extensions of bump-based technology, not a fundamental solution.

 

2. Hybrid Bonding: The Magic of ‘Direct Connection’

This led to a new idea: “Let’s just get rid of the bumps altogether!” That’s the start of hybrid bonding. The core concept is ‘direct connection.’ It’s a technology that bonds the copper pads and their surrounding insulating material directly to each other without any intermediate material, fusing the wafer or chip surfaces at an atomic level.

The process demands extreme precision. First, a process called CMP (Chemical-Mechanical Polishing) makes the wafer surface unbelievably smooth—so smooth that imperfections just a few atoms high are unacceptable. Next, the surface is activated with plasma to prepare it for bonding. Then, the two surfaces are aligned with incredible accuracy and brought into contact at room temperature, where they weakly stick together due to molecular forces. Finally, an annealing (heating) step allows the copper atoms and insulator molecules to diffuse into each other, forming a powerful and permanent bond.

⚠️ So what’s the big deal?
With no bumps, the connection pitch can be reduced to hundreds of nanometers. This means you can create millions of I/O connections per square millimeter. The shorter path drastically reduces electrical resistance and signal interference, leading to much higher speeds and significantly lower power consumption. The direct copper contact also improves heat dissipation, and the overall package becomes thinner.

 

3. A Nanometer-Scale War: The Challenges Ahead

While the benefits are clear, the reality of implementing it is a ‘war fought at the nanometer scale.’ The technical hurdles are immense.

  • Surface Flatness: Even a tiny bump just a few atoms high can cause the bond to fail. The surface needs to be far smoother than a billiard table. Managing the CMP process is key to achieving good yields.
  • Surface Cleanliness: A single nanoparticle can ruin the connection. Plasma dicing is preferred over traditional blade dicing because it generates fewer particles.
  • Alignment Accuracy: To connect pads with a pitch of a few hundred nanometers, the alignment error must be within tens of nanometers—a fraction of the width of a human hair. This requires real-time correction for tiny amounts of wafer warpage.
  • Copper Oxidation: Even a thin layer of oxidation on the copper surface can prevent a bond, making it one of the biggest headaches. Solutions involve bonding in a vacuum or coating the surface with less reactive metals.
  • Dielectric Material: Choosing the right insulator involves a trade-off between thermal expansion, bonding strength, and electrical properties, requiring careful selection of materials like SiO2, SiCN, or polymers.

 

4. W2W vs. D2W: The Two Faces of Hybrid Bonding

Hybrid bonding comes in two main flavors: Wafer-to-Wafer (W2W), ideal for mass production, and Die-to-Wafer (D2W), used for more complex, precise structures.

Category Wafer-to-Wafer (W2W) Die-to-Wafer (D2W)
Concept Bonds two entire wafers at once. Bonds individual, pre-tested good dies onto a wafer.
Features High throughput, relatively simple process. Can exclude defective dies, essential for heterogeneous integration.
Applications CMOS Image Sensors, 3D NAND. HBM, AI Accelerators, Logic (Intel Foveros, etc.).

The high-quality camera sensors in our smartphones are a success story for W2W. HBM, however, requires the D2W approach to stack multiple layers of pre-tested DRAM chips, similar to carefully constructing a skyscraper one floor at a time.

๐Ÿ’ก The Brutal Math of D2W Yield
D2W faces a challenge on a whole different level: the brutal math of cumulative yield. For example, if the yield for bonding one layer is 99%, the final yield after stacking 10 layers becomes 0.99^10, which is only about 90%. That 1% failure rate at each step results in a 10% final defect rate. As the number of layers increases, the yield drops exponentially, which is why pre-testing for Known Good Die (KGD) is absolutely critical.

 

5. Pushing Forward and a Final Question

Despite these challenges, the technology continues to advance. Active research in ‘low-temperature bonding’ aims to bring process temperatures below 150-200°C for heat-sensitive chips like DRAM. At the same time, engineers are tackling thermal stress issues through new materials, processes, and structural designs.

Hybrid bonding is now expanding beyond sensors and HBM to logic and HPC, with technologies like Intel's ‘Foveros’ and TSMC’s ‘SoIC.’ It is unquestionably the key that will unlock the next level of chip performance and density, but it remains a pinnacle of advanced technology with a mountain of challenges to overcome.

Recently, researchers successfully bonded completely different materials at room temperature, like silicon carbide (SiC) and silicon (Si). This makes you wonder: what if, in the future, we could bond any material to another with atomic precision? What new devices could be born? What unimagined systems could become possible? I’ll leave you with that question to ponder as we conclude our deep dive.

Frequently Asked Questions ❓

Q: What makes hybrid bonding better than traditional solder bumps?
A: The biggest differences are ‘connection density’ and ‘efficiency.’ By eliminating the physical bumps, you can create far more and shorter data pathways. This leads directly to faster processing speeds and lower power consumption, which is essential for high-performance chips used in AI.
Q: What’s the biggest reason it’s so hard to apply hybrid bonding to HBM?
A: It comes down to the ‘cumulative yield’ problem. HBM involves stacking many layers of DRAM (8, 12, or even 16), which requires the Die-to-Wafer (D2W) method. Because you're bonding one chip at a time, even a tiny chance of failure at each step multiplies, drastically lowering the probability of producing a perfect final product.
Q: Is hybrid bonding already being used in commercial products?
A: Yes, it's actively used in certain areas. The best example is the ‘CMOS Image Sensor (CIS)’ in smartphone cameras. Sony adopted Wafer-to-Wafer (W2W) hybrid bonding early on to dramatically improve camera performance. However, the D2W method needed for HBM is much more complex and is still in the R&D phase.

Can AI Be Your Paralegal? (Only if You Follow This 5-Step Verification Process)

  Blogging_CS · Sep 20, 2025 · 10 min read Generative AI promises to revo...