On June 17th our fourth AAIC - Applied AI Conference of the year took place, this time it was all about Applied AI in the semconductor industry. The conference was co-organized by AI Austria, Advantage Austria, and Silicon Alps.
Recently the semiconductor industry has gained much attention from AI startups, investors, and incumbents alike. On an abstract level, this development can be broken down into 2 areas: AI chips and improved chip manufacturing through AI.
Within just a few years the compute requirements to train state-of-the-art machine learning models have increased by a factor of 300.000. This fast rise is unlikely to slow down any time soon as very large models such as open AI‘s DALL-E or GPT-3 set new records for spent compute and parameters. Custom architectures to speed up the process of training and later-on increase the efficiency of deployment (inference) is, therefore, a logical consequence. At the same time, the „traditional“ semi-conductor industry is facing increasingly hard challenges when moving to the next production node, while maintaining profit margins and output. Like in many other industry verticals AI is playing a key role when it comes to improving the efficiency across the entire value chain - from design to manufacturing and logistics.
The AAIC Semiconductors is giving these exciting developments a stage, while also discussing the rekindled interest in EU-based chip manufacturing with leading European clusters and industry players.
Keynote #1: Applied AI at Infineon: Novel AI approaches and solutions in Infineon production
The morning session kicked off with a keynote by Dr.Thomas Altenmüller, Dr. Yao Yang, and Natalie Gentner from Infineon Technologies.
"With 1200 steps per wafer and 1000 involved tools, using AI is not an option but a necessity in semiconductor manufacturing"
The team provided insights on where across the value chain Infineon is already applying AI, as well as a deep-dive on the application of advanced methods like auto-encoders, transfer learning, and reinforcement learning.
Keynote #2: Low power in-sensor-ai applications in the charge domain
For our second talk we had a very special guest: Andreas Sibrai, the global VP of Design at AI Storm. Silicon Valley based AI Storm is at the forefront of edge AI and just recently has secured a funding round at a valuation of USD 250 Million. Unknown to many, their design team is located in Graz, Styria and led by Mr.Sibrai.
By including the AI logic directly in the sensor, AI Storm is pushing the concepts of edge AI and AI silicon to the limits. Especially when it comes to power consumption such in-sensor-AI approaches reduce power consumption by several orders of magnitude, without sacrificing speed or scalability.
One often-cited challenge for AI Chip startups is the lack of available development tools. By offering compatibility with existing frameworks such as TensorFlow AI Storm is tackling this entry barrier right at launch.
Keynote #3: Industrial Edge AI
For the last keynote in the morning session we invited Nico Teringl, the CEO of edge-AI startup Danube Dynamics.
In his talk Nico elaborated on the reasons behind edge AI such as low latency and data ownership and gave an overview of industrial applications.
Conclusion & Outlook
We hope that with AAIC Semiconductors we were able to demonstrate that AI in the semiconductor industry is not a still far-away concept but for many companies the new normal. We were especially happy to discover during the conference preparations that European companies and research institutions are very active in that field and hopefully the AAIC can contribute to more visibility and mindshare in the public discourse. Furthermore, we hope that when it comes to the implementation of the EU‘s digital agenda, AI silicon will get the attention it deserves, as Europe has a very strong base to build upon.