åǥÁö

[GT] Â÷¼¼´ë ÄÄÇ»ÅÍ ´º·Î¸ðÇÈÀ» À§ÇÑ °íÈ¿À² ¸Þ¸ð¸® ±â¼ú

ÇÇ¾î ¸®ºä(peer-reviewed) °úÇÐÀú³Î ¡¸¾îµå¹ê½ºµå ¸ÅÅ͸®¾ó½º(Advanced Materials)¡¹´Â ±â°è ÇнÀ(machine learning)À» ¹æÇØÇÏ´Â ÇÑ ¹®Á¦¸¦ ÇØ°áÇÏ´Â µ¥ µµ¿òÀÌ µÉ ¼ö ÀÖ´Â ºñÈֹ߼º ÄÄÇ»ÅÍ ¸Þ¸ð¸®(non-volatile computer memory, Àü¿øÀÌ ²÷¾îÁø »óÅ¿¡¼­µµ Á¤º¸¸¦ À¯ÁöÇÏ°í ÀÖ¾î Àü¿øÀÌ °ø±ÞµÇ¸é ´Ù½Ã ÀúÀåµÈ Á¤º¸¸¦ »ç¿ëÇÒ ¼ö ÀÖ´Â ±â¾ï ¸Åü)¸¦ ¸¸µå´Â »õ·Î¿î °³¹ß ¹æ¹ýÀ» ¼³¸íÇÏ°í ÀÖ´Ù. ÀÌ ¹æ¹ýÀº ¶ÇÇÑ À½¼º ÀνÄ, À̹ÌÁö ÇÁ·Î¼¼½Ì, ÀÚÀ² ÁÖÇà°ú °°Àº °¢Á¾ ±â¼úµéÀ» Çõ½ÅÇÒ ¼ö ÀÖ´Â ÀáÀç·Âµµ °®Ãß°í ÀÖ´Ù.



¾î¶² ½Å±â¼úÀÌ ¼¼»óÀ» ±ØÀûÀ¸·Î º¯È­½Ãų±î? ¼¼°è ÃÖ°íÀÇ ¿¬±¸¼Ò¿¡¼­ ³ª¿À´Â ³î¶ó¿î Çõ½ÅÀ» µ¶Á¡ ¼Ò°³ÇÕ´Ï´Ù.

 

 

ÇÇ¾î ¸®ºä(peer-reviewed) °úÇÐÀú³Î ¡¸¾îµå¹ê½ºµå ¸ÅÅ͸®¾ó½º(Advanced Materials)¡¹´Â ±â°è ÇнÀ(machine learning)À» ¹æÇØÇÏ´Â ÇÑ ¹®Á¦¸¦ ÇØ°áÇÏ´Â µ¥ µµ¿òÀÌ µÉ ¼ö ÀÖ´Â ºñÈֹ߼º ÄÄÇ»ÅÍ ¸Þ¸ð¸®(non-volatile computer memory, Àü¿øÀÌ ²÷¾îÁø »óÅ¿¡¼­µµ Á¤º¸¸¦ À¯ÁöÇÏ°í ÀÖ¾î Àü¿øÀÌ °ø±ÞµÇ¸é ´Ù½Ã ÀúÀåµÈ Á¤º¸¸¦ »ç¿ëÇÒ ¼ö ÀÖ´Â ±â¾ï ¸Åü)¸¦ ¸¸µå´Â »õ·Î¿î °³¹ß ¹æ¹ýÀ» ¼³¸íÇÏ°í ÀÖ´Ù. ÀÌ ¹æ¹ýÀº ¶ÇÇÑ À½¼º ÀνÄ, À̹ÌÁö ÇÁ·Î¼¼½Ì, ÀÚÀ² ÁÖÇà°ú °°Àº °¢Á¾ ±â¼úµéÀ» Çõ½ÅÇÒ ¼ö ÀÖ´Â ÀáÀç·Âµµ °®Ãß°í ÀÖ´Ù.

»÷µð¾Æ ±¹¸³¿¬±¸¼Ò(Sandia National Laboratories)¿Í ¹Ì½Ã°Ç ´ëÇÐ(University of Michigan)ÀÇ ¿¬±¸ÆÀÀº ÄÄÇ»ÅÍ Ä¨¿¡ ´õ ¸¹Àº ÇÁ·Î¼¼½Ì ¼º´ÉÀ» ´ã´Â »õ·Î¿î ¹æ¹ý¿¡ °üÇÑ »ó¼¼ÇÑ ³í¹®À» ¹ßÇ¥Çß´Ù. À̵éÀº ¿¡³ÊÁö È¿À²ÀÌ ³ôÀº ¡®±â°è Ãß·Ð(machine inference)¡¯ ÀÛµ¿À» °¡´ÉÄÉ ÇÏ´Â ¾Æ³¯·Î±× ¸Þ¸ð¸® ÀåÄ¡¿¡ ÁÖÅÿë ÆäÀÎÆ®¿¡¼­ ¹ß°ßµÇ´Â ÀÏ¹Ý ¹°ÁúÀ» È°¿ëÇÏ´Â ¹æ¹ýÀ» Àû¿ëÇß´Ù.

 

»êÈ­ ƼŸ´½(titanium oxide)Àº ¼¼°è¿¡¼­ °¡Àå ÀϹÝÀûÀÎ ¹°Áú Áß ÇϳªÀÌ´Ù. ¿ì¸®°¡ ±¸¸ÅÇÏ´Â ¸ðµç ÆäÀÎÆ®¿¡´Â »êÈ­ ƼŸ´½ÀÌ µé¾îÀÖ´Ù. ÀÌ ¹°ÁúÀº Àú·ÅÇÏ°í µ¶¼ºÀÌ ¾øÀ¸¸ç, ÀÏÁ¾ÀÇ »êÈ­¹°ÀÌ´Ù. ±×¸®°í ÀÌ ¹°Áú¿¡¼­ ¸î °³ÀÇ »ê¼Ò ¿øÀÚ¸¦ Á¦°ÅÇϸé, ¿ì¸®´Â ¼ÒÀ§ »ê¼Ò °ø°ø(oxygen vacancy)À¸·Î ºÒ¸®´Â °ÍÀ» ¸¸µé ¼ö ÀÖ´Ù. ¹àÇôÁø ¹Ù¿Í °°ÀÌ, »ê¼Ò °ø°øÀÌ ¸¸µé¾îÁ³À» ¶§, »êÈ­ ƼŸ´½Àº Àü±â Àüµµ¼ºÀ» °®°Ô µÈ´Ù.

 

ÀÌ·¯ÇÑ »ê¼Ò °ø°øÀº ÀÌÁ¦ Àü±â µ¥ÀÌÅ͸¦ ÀúÀåÇÒ ¼ö ÀÖ¾î, °ÅÀÇ ¸ðµç ÀåÄ¡¿¡ ´õ ¸¹Àº ÄÄÇ»Æà ´É·ÂÀ» ºÎ¿©ÇÑ´Ù. ¿¬±¸ÆÀÀº »êÈ­ ƼŸ´½À¸·Î ÄÚÆÃµÈ ÄÄÇ»ÅÍ Ä¨À» È­¾¾ 302µµ ÀÌ»óÀ¸·Î °¡¿­ÇÏ¿©, °ø°øÀ» ¸¸µé±â À§ÇÑ Àü±âÈ­ÇÐÀ» ÅëÇØ ÀÌ ¹°Áú¿¡¼­ »ê¼Ò ºÐÀÚµéÀ» ÀϺΠºÐ¸®ÇÔÀ¸·Î½á »ê¼Ò °ø°øÀ» ¸¸µé¾ú´Ù. ÀÌÈÄ ¿­À» ½ÄÈù ÀÌ ¹°ÁúÀº ¿ì¸®°¡ ÇÁ·Î±×·¥ÇÑ ¸ðµç Á¤º¸¸¦ ÀúÀåÇÒ Áغñ¸¦ °®Ãß°Ô µÇ¾ú´Ù.

 

¿À´Ã³¯, ÄÄÇ»ÅÍ´Â ÀϹÝÀûÀ¸·Î ÇÑ °÷¿¡ µ¥ÀÌÅ͸¦ ÀúÀåÇÏ°í ´Ù¸¥ °÷¿¡ ±× µ¥ÀÌÅ͸¦ ÇÁ·Î¼¼½ÌÇÔÀ¸·Î½á ÀÛµ¿ÇÑ´Ù. ÀÌ°ÍÀº ÄÄÇ»ÅÍ°¡ ÇÑ °÷¿¡¼­ ´Ù¸¥ °÷À¸·Î µ¥ÀÌÅ͸¦ ²÷ÀÓ¾øÀÌ Àü¼ÛÇØ¾ß ÇÑ´Ù´Â Àǹ̷Î, ¿¡³ÊÁö¿Í ÄÄÇ»Æà ¼º´ÉÀ» ³¶ºñÇÏ´Â °ÍÀÌ´Ù.

 

¿¬±¸ÆÀÀ» À̲ô´Â ¼ö¼® ¿¬±¸¿øÀº ¹ßÇ¥µÈ ÇÁ·Î¼¼½º°¡ ÄÄÇ»ÅÍ ÀÛµ¿ ¹æ½ÄÀ» ¿ÏÀüÈ÷ ¹Ù²Ü ¼ö ÀÖ´Â ÀáÀç·ÂÀ» ¾î¶»°Ô °¡Áö°í ÀÖ´ÂÁö ¼³¸íÇß´Ù. ¿¬±¸ÆÀÀÌ ÇÑ ÀÏÀº µ¿ÀÏÇÑ Àå¼Ò¿¡¼­ ÇÁ·Î¼¼½Ì°ú ÀúÀåÀÌ ÀÌ·ïÁö°Ô ÇÑ °ÍÀ̾ú´Ù. ¶ÇÇÑ ÀÌ°ÍÀº ¿¹Ãø °¡´ÉÇÏ°í ¹Ýº¹ °¡´ÉÇÑ ¹æ½ÄÀ¸·Î ¼öÇà µÉ ¼ö ÀÖ´Ù.

 

¿¬±¸¿øµéÀº ÀÌ·¯ÇÑ »ê¼Ò °ø°øÀÇ È°¿ëÀ» ÇöÀç ±â°è ÇнÀÀÌ Á÷¸éÇØ ÀÖ´Â °Å´ëÇÑ Àå¾Ö¹°À» ±Øº¹ÇÏ´Â ÇϳªÀÇ ¹æ¹ýÀ¸·Î º¸°í ÀÖ´Ù. ±× Àå¾Ö¹°À̶õ ¹Ù·Î Àü·Â ¼Ò¸ð(power consumption)ÀÌ´Ù.

 

±â°è ÇнÀÀ» ¼öÇàÇÏ·Á¸é ¸·´ëÇÑ ¿¡³ÊÁö°¡ ¼Ò¿äµÇ´Âµ¥, ±â°è°¡ µ¥ÀÌÅÍ ¾ÕµÚ·Î À̵¿Çϸ鼭 Àü·ÂÀ» ¼ÒºñÇϱ⠶§¹®ÀÌ´Ù. ¼ö¼® ¿¬±¸¿øÀº ÀÌ·¸°Ô ¸»ÇÑ´Ù.

 

¡°ÀÚÀ² ÁÖÇà ÀÚµ¿Â÷°¡ ¿¹·Î µç´Ù¸é, ¿îÀü¿¡ ´ëÇÑ °áÁ¤À» ³»¸®´Â µ¥ ¸¹Àº ¿¡³ÊÁö¸¦ ¼ÒºñÇÏ¿© ¸ðµç ÀÎDz(input) Á¤º¸¸¦ ó¸®ÇÕ´Ï´Ù. ÄÄÇ»ÅÍ Ä¨ÀÇ ´ëü Àç·á¸¦ ¸¸µé ¼ö ÀÖ´Ù¸é, Á¤º¸¸¦ º¸´Ù ´õ È¿À²ÀûÀ¸·Î ó¸®ÇÏ°í ¿¡³ÊÁö¸¦ Àý¾àÇÏ°í ´õ ¸¹Àº µ¥ÀÌÅ͸¦ ó¸®ÇÒ ¼ö ​​ÀÖ½À´Ï´Ù. ÇöÀçÀÇ ÈÞ´ëÆùÀ» »ý°¢ÇØ º¾½Ã´Ù. À½¼º ¸í·ÉÀ» ³»¸®°í ½Í´Ù¸é, À½¼ºÀ» µè´Â ÄÄÇ»ÅÍÀÇ Áß¾Ó Çãºê·Î ¸í·ÉÀ» Àü¼ÛÇÑ ´ÙÀ½ ÀüÈ­±â¿¡ ¼öÇàÇÒ ÀÛ¾÷À» ¾Ë¸®´Â ½ÅÈ£¸¦ ´Ù½Ã º¸³»´Â ³×Æ®¿öÅ©¿¡ ¿¬°áÇØ¾ß ÇÕ´Ï´Ù. ±×·¯³ª ÀÌ »õ·Î¿î ÇÁ·Î¼¼½º¸¦ ÅëÇØ À½¼º ÀÎ½Ä ¹× ±âŸ ±â´ÉÀÌ ÈÞ´ëÆù ³»¿¡¼­ ¹Ù·Î ÀÌ·ç¾îÁú ¼ö ÀÖ½À´Ï´Ù.¡±

 

¿¬±¸ÆÀÀº ÇöÀç ¿©·¯ ÇÁ·Î¼¼½º¸¦ °³¼±ÇÏ°í ´õ Å« ±Ô¸ð·Î Å×½ºÆ®¸¦ ¼öÇà ÁßÀÌ´Ù.

 

References
Advanced Materials
, September 22, 2020, ¡°Filament‐free bulk resistive memory enables deterministic analog switching,¡± Yiyang Li, et al. © 2020 John Wiley & Sons, Inc. All rights reserved.

 

To view or purchase this article, please visit:
https://onlinelibrary.wiley.com/doi/10.1002/adma.202003984
Filament‐Free Bulk Resistive Memory Enables Deterministic Analogue Switching - Li - 2020 - Advanced Materials - Wiley Online Library

The peer-reviewed journal Advanced Materials describes the development of a new method to make non-volatile computer memory which may help solve a problem that has been holding back machine learning. And it has the potential to revolutionize technologies like voice recognition, image processing, and autonomous driving.

 

A team from Sandia National Laboratories and the University of Michigan published a paper detailing a new method that will imbue computer chips which power machine-learning applications with more processing power by using a common material found in house paint in an analog memory device which will enable highly energy-efficient ¡°machine inference¡± operations.

 

Titanium oxide is one of the worlds¡¯ most common materials. Every painting you buy has titanium oxide in it.  It¡¯s cheap and nontoxic. It¡¯s an oxide, and if you take a few oxygen atoms out, you create what are called oxygen vacancies. As it turns out, when you create oxygen vacancies, you make titanium oxide electrically conductive.

 

Those oxygen vacancies can now store electrical data, giving almost any device more computing power. The team created the oxygen vacancies by heating a computer chip with a titanium oxide coating above 302 degrees Fahrenheit and then separating some of the oxygen molecules from the material using electrochemistry to create vacancies. When it cooled off, it was ready to store any information you program it with.

 

Right now, computers generally work by storing data in one place and processing that data in another place. That means computers have to constantly transfer data from one place to the next, wasting energy and computing power.

 

The lead researcher explained how the process has the potential to completely change how computers work.  What the team did was make the processing and the storage at the same place. What¡¯s new is that this can be done in a predictable and repeatable manner.

 

The researchers see the use of oxygen vacancies as a way to help machine learning overcome a big obstacle holding it back right now: power consumption.

 

Doing machine learning takes a lot of energy because the machine is moving data back and forth causing power consumption. The lead researcher says, ¡°If you have autonomous vehicles, making decisions about driving consumes a large amount of energy to process all the inputs.  If we can create an alternative material for computer chips, they will be able to process information more efficiently, saving energy and process a lot more data. Think about your cell phone. If you want to give it a voice command, you need to be connected to a network that transfers the command to a central hub of computers that listens to your voice and then sends a signal back telling your phone what to do. Through this new process, voice recognition and other functions could happen right in your phone.¡±

 

The team is now working on refining several processes and testing the method on a larger scale.

 

References
Advanced Materials
, September 22, 2020, ¡°Filament‐free bulk resistive memory enables deterministic analog switching,¡± Yiyang Li, et al. © 2020 John Wiley & Sons, Inc. All rights reserved.

 

To view or purchase this article, please visit:
https://onlinelibrary.wiley.com/doi/10.1002/adma.202003984
Filament‐Free Bulk Resistive Memory Enables Deterministic Analogue Switching - Li - 2020 - Advanced Materials - Wiley Online Library