The impact of AI on the semiconductor industry is not limited to driving demand; it extends to changing the way the entire industry operates. According to the vision presented by John Kibarian, CEO of PDF Solutions, AI is reshaping design tools, data models, and collaboration methods used in the manufacturing and testing of silicon chips.
The Energy Efficiency Dilemma and Accelerated Growth
Projections indicate massive growth in device revenues, reaching approximately $2.3 trillion by 2030, with 58% of this increase coming from data centers and servers. This growth translates into a massive increase in the capacity of AI data centers, which could reach 160 GW by 2030.
However, building fabs and increasing production capacity is not the only solution; the real challenge lies in energy efficiency. Hyperscalers are now moving toward designing custom chips that focus primarily on energy efficiency, changing the game from merely providing high performance to achieving maximum computing per watt.
Data Efficiency: Overcoming the Obstacles of Cost and Complexity
The manufacturing process faces a technical challenge in the form of massive data volumes; large fabs possess petabytes of data across multiple domains. Here, the "token cost" problem in Large Language Models (LLMs) emerges. Running an LLM for every data analysis process can be extremely expensive and energy-inefficient.
To solve this dilemma, a new strategy is emerging that relies on using language models to write Python workflows instead of performing the analysis directly. This approach offers significant technical advantages:
- Cost reduction: Instead of consuming thousands of tokens in every analysis, the workflow is created once and then executed with minimal energy cost.
- Reproducibility: Engineers can store a successful workflow and use it later, ensuring consistent results.
- Data security: This model allows for air-gapping sensitive customer data from language models, where code is run locally without the need to send raw data to the cloud.
Supply Chains and Agentic Collaboration
The semiconductor industry is characterized by complex supply chains, where a chip passes through multiple stages in different facilities around the world (from wafer manufacturing to advanced packaging and testing). With the advent of AI, collaboration is shifting from mere human communication to "agentic collaboration," where systems and software agents exchange data and algorithms automatically and securely.
This shift requires ultra-secure communication networks, such as those provided by secureWISE technologies, to ensure the transfer of algorithms and data between suppliers and fabs without compromising intellectual property or cybersecurity.
What Does This Mean for Manufacturers and Developers?
This transformation imposes a new reality that requires companies to adapt to several fundamental changes:
- Shift toward custom design: The need for chips specifically designed for certain applications will increase, meaning smaller production volumes and greater product diversity.
- Integration of design and manufacturing: Fabless companies can no longer ignore manufacturing software; co-optimization between design and manufacturing processes has become a necessity to ensure yield in advanced technology nodes (such as sub-7nm).
- Reliance on intelligent automation: Engineers will increasingly rely on analytical tools that use AI to generate code and manage big data, raising the expectations for human talent to focus on context and verification rather than routine operations.