CONNECT WITH US

Advanced vision AI eliminates potential risks in cleanroom environments for semiconductor and panel manufacturers

Press release 0

Industries such as semiconductors and panel manufacturing, which the world depends on, implement cleanrooms to prevent contaminants such as microorganisms, suspended particles, volatile chemical gases, or dust that can decrease yield rates and result in considerable losses in the production process.

It's undeniable that the high degree of cleanliness expected by the above industries cannot be solely achieved through isolated spaces such as cleanrooms, but must also rely on personnel's compliance to good cleaning practices. This means that personnel must wear protective clothing/hats/shoes before entering cleanrooms and raise both arms when walking through air shower tunnels, so that clean air may remove dust particles attached to these items. While these rules are in place, sometimes colleagues may fail to properly raise both arms, walk quickly through the air shower tunnel, fail to rotate properly, or spend insufficient time in the air shower tunnel, affecting the product's yield rate.

In order to eliminate these circumstances, more and more businesses are requiring in house development teams to work on AI surveillance models to identify whether personnel's behaviors in air shower tunnels are compliant with existing rules; in the event of a violation, a linked warning mechanism can instantly request personnel to undergo comprehensive cleaning procedures. Due to the complex development of these models, the completion of many projects has been put on hiatus. Luckily, the NEON AI Smart Camera developed by ADLINK has provided timely assistance and helped revive numerous projects towards successful landings.

AI Engineers Don't Understand Imaging? Potential Delays to Program Development

Chia-Wei Yang, Director at ADLINK's Edge Vision Business Center stated that the majority of factory AI engineers have backgrounds in automation and have only changed their fields in recent years; while they have undoubtedly honed their ability in AI over many years, they do not understand imaging in the sense that they don't know how to select lenses, light sources, or understand camera settings. As such, their efforts in developing image capture programs will naturally encounter a challenging learning threshold.

However, the difficulties don't stop here. Challenges await at each stage, including environment restrictions (such as the inability to drill holes in cleanrooms, minimizing cabling, space restrictions, and the need for fanless designs to prevent dust accumulation), ensuring system reliability, and real-time DIO.

Tsai-Pan Lin, Solution Engineer Director at ADLINK's Edge Vision Business Center, pointed out that this would not be an easy task due to the development processes that these AI engineers are accustomed to. Since it is impossible for them to develop programs directly in cleanroom environments, they need to record videos and use them as the basis for writing testing programs, which require integrating posture judgment logic to identify whether personnel's actions should be deemed as "Pass" or "NG." Once testing is confirmed to be error free, they have to go through the time consuming process of coming up with methods to integrate edge devices and cameras, and then deploy them into designated spaces.

Taking Advantage of the EVA SDK to Land Projects within Two Weeks

After implementing the NEON AI Smart Camera, AI teams will only need to develop one inference program to utilize the clear function block guide provided by EVA SDK (the NEON AI Camera's built-in edge vision analytics software development kit). This allows developers to give a tangible form to their logic, swiftly introduce it into programs, and rapidly toggle image sources from video to camera. In other words, one program can meet testing and subsequent real-life scenario flattening requirements. Generally speaking, users can utilize visualization tools in the built-in EVA IDE of EVA SDK to convert complex development processes into intuitive flow charts, and then place function blocks in order and achieve project landing within two weeks.

As for environment restrictions, the NEON AI Smart Camera provides massive assistance primarily because it is an All-in-One system with an integrated camera and computer that only requires one network and a power cable to meet businesses' expectations by "reducing cabling to a minimum." Furthermore, NEON features ease of installation and supports the VESA standard installation interface, so that users can obtain various mechanisms available in the market to complete setup operations, thereby eliminating the necessity to drill holes. More importantly, NEON utilizes a fanless design, which eliminates concerns regarding dust accumulation.

In the past, AI engineers utilized development boards and webcams readily available to users for in house developed programs, which often suffered from unknown factors that resulted in poor inference reliability and images being bypassed. Also, additional efforts were required to develop I/O control programs to immediately trigger warning mechanisms such as three-color indicators and voice warnings. After implementing NEON, pre-operations such as image capture, verification, and adding algorithms can be completed and quickly setup within two weeks to provide stable image capture without bypass as well as stable inference time. The addition of EVA SDK's built-in I/O control plug-in also allow AI engineers to build I/O control logic through low-code development methods.

The clear benefits provided by NEON with EVA SDK can offer massive value to businesses. Based on responses from various customers, AI surveillance has increased personnel compliance; they no longer disregard cleaning regulations and undergo proper cleaning in air shower tunnels, decreasing dust by a massive 80% and significantly eliminating factors that affect yield rates.

Chia-Wei Yang stated that ADLINK is pleased with the positive customer feedback and expects to develop additional tools "that will lower learning curves," so that customers can accelerate their implementation of vision AI applications; these little tools will appear on the inference side and eventually for the training side. Additionally, ADLINK will continue to organize in-depth AI academy courses to cultivate even more talented individuals capable of developing advanced vision AI solutions.

For more information, please visit: https://www.adlinktech.com/en/EVA_SDK_ADLINK_Edge_AI_Vision_Analytics.

NEON AI Smart Camera developed by ADLINK has provided timely assistance and helped revive numerous projects towards successful landings.

NEON AI Smart Camera developed by ADLINK has provided timely assistance and helped revive numerous projects towards successful landings.