»ó´Ü¿©¹é
HOME Çлý±âÀÚ´Ü
»ï¼ºÀüÀÚ, ¡®¹ÝµµÃ¼ ºñÀü 2030¡¯ À§ÇØ ¡®AIÀÇ ³ú¡¯ NPU¿¡ Àü·ÂÅõ±¸
ÃÖ½ÅÇõ °­³²Æ÷½ºÆ® ÀÎÅϱâÀÚ | ½ÂÀÎ 2019.09.18 18:50

»ï¼ºÀüÀÚ°¡ ±Û·Î¹ú 1À§¸¦ ¸ñÇ¥ÇÑ ‘¹ÝµµÃ¼ ºñÀü 2030’ ´Þ¼ºÀ» À§ÇØ ÀΰøÁö´É(AI) »ê¾÷ÀÇ ÇÙ½ÉÀÎ NPU(Neural Processing Unit-½Å°æ¸Áó¸®ÀåÄ¡) ±â¼ú °³¹ß¿¡ º»°Ý ³ª¼¹´Ù. »ï¼ºÀüÀÚ ½Ã½ºÅÛLSI »ç¾÷ºÎÀåÀÎ °­Àο± »çÀåÀº Áö³­ 6¿ù 18ÀÏ ¼­¿ï ÅÂÆò·Î ºôµù¿¡¼­ ¿­¸° ±âÀÚ°£´ãȸ¿¡¼­ “AI »ýÅ°èÀÇ ÇÙ½ÉÀÎ NPU »ýÅ°è´Â ¾ÆÁ÷ ±â¾÷ °£ ±â¼ú¼öÁØÀ» ºñ±³ÇÒ ¸¸ÇÑ ¼öÁØÀÌ ¾Æ´Ñ (Ãʱâ) ´Ü°è”¶ó¸ç “°ü·Ã »ç¾÷À» ¼±Á¡ÇØ ½Ã½ºÅÛ ¹ÝµµÃ¼ 1À§¸¦ ´Þ¼ºÇÏ°Ú´Ù”°í ¹àÈ÷°í, “¼øÂ÷ÀûÀ¸·Î NPU žÀçÈ®´ë¸¦ °­È­ÇÒ °Í”À̶ó¸ç ¹àÇû´Ù.

µÚÀÌ¾î »ï¼ºÀüÀÚ Á¾ÇÕ±â¼ú¿ø Ȳ¼º¿ì ºÎ¿øÀåÀº "NPUÀÇ ÀåÁ¡Àº ¿Âµð¹ÙÀ̽º AI (On-Device AI)±¸Çö À¯Çü¼º°ú ¼öÁýµÈ µ¥ÀÌÅ͸¦ Ŭ¶ó¿ìµå·Î º¸³»Áö ¾Ê°í ±â±â ÀÚü¿¡¼­ ó¸®ÇÒ ¼ö ÀÖ´Â ±â´É°ú °³ÀÎÁ¤º¸ º¸È£ ±×¸®°í ÃÖÀûÇÑ ÀÀ´ä¼Óµµ ¹× Àü·Â ¼Ò¸ð"¶ó¸ç, NPU°¡ ¿Ö ÀΰøÁö´É(AI) ¹ÝµµÃ¼ ½ÃÀåÀÇ ÇÙ½ÉÀÌ µÇ°í ÀÖ´ÂÁö¸¦ ¼³¸íÇß´Ù. NPU´Â‘Àΰø ´º·²’³×Æ®¿öÅ©¸¦ ±¸ÇöÇϱâ À§ÇØ ÀΰøÁö´ÉÀÇ ÇÙ½ÉÀÎ µö·¯´×(Deep Learning) ¾Ë°í¸®Áò ¿¬»êÀ» Àû¿ëÇÑ ÇÁ·Î¼¼¼­ À¯´Ö(½Å°æ¸Á ó¸® ÀåÄ¡)ÀÌ´Ù. 2010³â Ä÷ÄÄ(Qualcomm)ÀÌ Ã³À½À¸·Î ¼Ò°³ÇÑ ÀÌ ÀΰøÁö´É ĨÀº Àΰ£ÀÇ ³ú ½Å°æ¸ÁÀ» ¸ð¹æÇØ‘½º½º·Î ÆÇ´ÜÇÏ´Â ´É·Â’À» °®Ãß°í ÀÖ´Ù. ƯÈ÷, »ç¶÷ÀÇ ³ú¿¡ ÀÖ´Â ´º·±Ã³·³ ¼ö¸¹Àº Äھ È°¿ëÇÑ NPU´Â µ¿½Ã¿¡ ´Ù¾çÇÑ ´ë¿ë·® µ¥ÀÌÅ͸¦ ó¸® ¹× ºÐ¼®À» ÇÒ ¼ö ÀÖ´Â ½º¸¶Æ®ÇÑ ±â´ÉÀ» Æ÷ÇÔÇÏ°í ÀÖ´Ù.

»ï¼ºÀüÀÚ´Â 2012³âºÎÅÍ ½Ã½ºÅÛ LSI»ç¾÷ºÎ¿Í Á¾ÇÕ±â¼ú¿øÀ» µ¿¿øÇØ ¹Ì·¡ AI½ÃÀå¿¡ ÁÖµµ±ÇÀÌ µÉ ‘Â÷¼¼´ë NPU ±â¼ú’°³¹ßÀ» Áö¼ÓÇØ ¿À°í ÀÖ¾ú´Ù.±× ù ½ÇÀû¹°·Î´Â ¿ÃÇØ °ø°³µÈ °¶·°½Ã S10¿¡ ÃÖÃÊ·Î NPU¸¦ žÀçÇÑ ½ÅÇü ¸ð¹ÙÀÏ ÇÁ·Î¼¼¼­ ¿¢½Ã³ë½º 9820°¡ ÀÖ´Ù. ÀüÀÛ ¿¢½Ã³ë½º 9810 ¸ðµ¨¿¡ ºñÇØ ÀΰøÁö´É ¿¬»ê´É·ÂÀÌ ¾à 7¹è, ÄÚ¾î ¼º´ÉÀº ¾à 20%, Àü·ÂÈ¿À²Àº 40% Çâ»óµÆ´Ù.

ƯÈ÷ NPUÀÇ È°¿ëµµ´Â Ä«¸Þ¶ó ¼º´É ¾÷±×·¹À̵忡¼­ ºñÁßÀÌ ÄÇ´Ù. »ï¼ºÀüÀÚ´Â À̹ø ¾÷±×·¹À̵带 ÅëÇØ ±×µ¿¾È‘ÃàÀûÇÑ µð½ºÇ÷¹ÀÌ, Ä«¸Þ¶ó ±â¼úÀ» ¹ÙÅÁÀ¸·Î Â÷º°È­µÈ µðÀÚÀΰú ±â´ÉÀ» ¼±º¸¿´À¸¸ç, ÃÔ¿µÇÒ ¶§¿¡ ÇÇ»çüÀÇ ÇüÅÂ?Àå¼Ò/ÁÖº¯ ¹à±â µî ȯ°æÀ» ¼ø°£ÀûÀ¸·Î °¨ÁöÇÏ°í, ÃÖÀû °ªÀ» ÀÚµ¿ÀûÀ¸·Î ¼³Á¤ÇÔÀ¸·Î½á ÃÖ»óÀÇ À̹ÌÁö¿Í ¿µ»óÈ­Áú Çâ»óÀ» ÀçÁ¶¸íÇß´Ù.

»ï¼ºÀüÀÚ´Â ÇâÈÄ NPU°¡ ½º¸¶Æ®Æù¿¡¸¸ Àû¿ëµÇ´Â °Í»Ó¸¸ ¾Æ´Ï¶ó ÀÚÀ²ÁÖÇàÂ÷¸¦ Æ÷ÇÔÇØ IT ÀüºÐ¾ßÀÇ IoT¿Í SoC (System on Chip)¿¡ È°¿ëµÇ¾î »ç¾÷ ¹üÀ§¸¦ ³ÐÈú ÀΰøÁö´É ĨÀ̶ó È®½ÅÇß´Ù. ÀÌ¿¡ Ȳ¼º¿ì ºÎ¿øÀåÀº "2025³â¿¡´Â ÀÚÀ¯ ´ëÈ­Çü À½¼ººñ¼­ µîÀÌ °¡´ÉÇϸç 2030³â ÀÌÈÄ¿¡´Â »ç¶÷°ú °°ÀÌ ÇൿÇϴ‘ÈÞ¸Ó³ëÀ̵咵µ ³ª¿Ã ¼ö ÀÖÀ» °Í"À̶ó°í Àü¸ÁÇß´Ù.

NPU ¹ÝµµÃ¼ ½ÃÀå ¼±Á¡À» À§ÇØ ÇöÀç »ï¼ºÀüÀÚ´Â 200¿© ¸íÀÇ ÀηÂÀ» 2030³â±îÁö 10¹è ÀÌ»óÀ¸·Î È®´ëÇÒ °èȹÀ̸ç, IT °³¹ßÀÚ ÀÎÀçÈ®º¸ ¹× ±Û·Î¹ú ¿¬±¸ ±â°ü Çù·Â¿¡ ÁýÁßÇÑ´Ù´Â µµ¾àÀ» ¼±º¸¿´´Ù.

ÇÑÆí »ï¼ºÀüÀÚ´Â Áö³­ 5¿ù ¼¼°èÀû ¼®ÇÐÀÎ ¿ä½´¾Æº¥Áö¿À(YoshuaBengio) ±³¼ö¸¦ Áß½ÉÀ¸·Î ¸óÆ®¸®¿Ã´ë(University of Montreal), ¸Æ±æ´ë(McGill University) ¿¬±¸Áø°ú °³¹ßÀÚ µî°úÀÇ Çù·ÂÀ» ½ÃÀÛÇßÀ¸¸ç, µö·¯´× Àü¹® ¿¬±¸ È®ÀåÀ» À§ÇØ Ä³³ª´Ù ¹Ð¶ó¿¬±¸¼Ò¿Í °áÇÕÇß´Ù. ¶ÇÇÑ 2017³âºÎÅÍ ´º·²ÇÁ·Î¼¼½Ì¿¬±¸¼¾ÅÍ(NPRC, Neural Processing Research Center)¿Í ¿¬ÇÕÇØ ±¹³» ´ëÇеé°úÀÇ Áö¼Ó Çù·ÂÀ» ÃßÁøÇÔÀ¸·Î½á‘¹ÝµµÃ¼ ºñÀü 2030’ ´Þ¼º¿¡ Àü·Â Åõ±¸ÇÏ°í ÀÖ´Ù.

 

Samsung Electronics Accelerating ‘Semiconductor Vision 2030’

 

Samsung Electronics, a predominant leader in system-semiconductor technologies, has revealed their plan to develop a neural processing unit (NPU), a new core of artificialIntelligence (AI) industry, in pursuit of reaching ‘Semiconductor Vision 2030’ goal. Earlier, Samsung Electronics held a briefing on the NPU project at Seoul Taepyung-ro Building on June 18. “The NPU, the core of the AI ecosystem, is not yet at the level of comparable technology levels between rival companies. But we will gradually preoccupy the market by becoming No. 1 in advanced system semiconductor,” said Kang In-yeop, president of company’s System LSI Division.

Subsequently Hwang Sung-woo, the vice chairman of Samsung Advanced Institute of Technology, explained why NPU is the core of AI market. “The great advantages,”he said, “include the flexible On-Device AI implementation, the ability to process the collected data on the device itself without sending it to the cloud, privacy, and optimal response time plus low power consumption”.

The NPU is an AI-based processor unit functioning by ‘Deep Learning’ algorithm computation that is to ultimately implement in ‘neural network’ processing device. First introduced by Qualcomm in 2010, NPU is known as a human brain-like AI chip and has the ability to judge on its own. Particularly, utilizing many cores like neurons in the human brain, NPU contains smart features that can process and analyze a large amount of data ‘simultaneously’.

Since 2012, Samsung Electronics has been in progress of developing primary NPU technology through the collaboration of System LSI Division and Advanced Institute of Technology. The first NPU equipped outcome was the new mobile processor Exynos 9 (9820) which was implemented in Galaxy S10 this year. Compared to the previous Exynos 9810 model, this NPU equipped model has proven the substantial enhancement such as 7 times at computing power, 20% at core performance, and 40% at power efficiency.
Especially the utilization of NPU made a great contribution on camera performance upgrade. Samsung Electronics has unveiled ‘different design and function based on accumulated display and camera technology’ through this upgrade. When shooting, the camera instantly detects the 'environment such as the shape, location, and surrounding brightness' of the subject, and automatically fixes the optimum value to produce the best image and image quality.

Samsung Electronics was convinced that NPU will not only be applied to smartphones but also artificial intelligence chips that will be used in all areas of IT such as in IoT (Internet on Things) and SoC (System on Chip) including autonomous vehicles. "We will be able to have a free interactive voice assistant in 2025 andin 2030possibly humanoid that will behave like a human,”said Hwang Sung-woo.

In order to dominate the NPU semiconductor market in advance, Samsung Electronicsplans to create over 2,000 high-caliber talent worldwide by 2030, which is about 10 times the current headcount,concentrating on securement of IT developer talent and cooperation with global research institutes.

Meanwhile in May, Samsung Electronics began collaborating with researchers and developers at the Montreal University and McGill University, including world-renowned researcher Professor YoshuaBengio. And they combined concerted operation with Canadian Mila Research Institute for specializing ‘Deep Learning’. Furthermore, since 2017, the company has been running full power to achieve ‘Semiconductor Vision 2030’ by partnering with the Neural Processing Research Center (NPRC) and promoting continuous cooperation with domestic universities.

ÃÖ½ÅÇõ °­³²Æ÷½ºÆ® ÀÎÅϱâÀÚ  webmaster@ignnews.kr

<ÀúÀÛ±ÇÀÚ © °­³²Æ÷½ºÆ®, ¹«´Ü ÀüÀç ¹× Àç¹èÆ÷ ±ÝÁö>

ÃÖ½ÅÇõ °­³²Æ÷½ºÆ® ÀÎÅϱâÀÚÀÇ ´Ù¸¥±â»ç º¸±â
iconÀαâ±â»ç
½Å¹®»ç¼Ò°³¤ý±â»çÁ¦º¸¤ý±¤°í¹®ÀǤýºÒÆí½Å°í¤ý°³ÀÎÁ¤º¸Ãë±Þ¹æħ¤ýû¼Ò³âº¸È£Á¤Ã¥¤ýÀ̸ÞÀϹ«´Ü¼öÁý°ÅºÎ
¼­¿ï½Ã °­³²±¸ ¼±¸ª·Î 704, 10Ãþ 593È£(û´ãµ¿, û´ãº¥Ã³ÇÁ¶óÀÚ)  |  ´ëÇ¥ÀüÈ­ : 02)511-5877   |  ¹ßÇàÀÏÀÚ : 1995³â 4¿ù 6ÀÏâ°£
µî·ÏÀÏÀÚ : 2018³â 2¿ù 28ÀÏ  |  µî·Ï¹øÈ£ : ¼­¿ï ¾Æ 04996  |  È¸Àå : Á¶¾çÁ¦  |   ¹ßÇàÀÎ : Á¶ÀÎÁ¤  |  ÆíÁýÀÎ : Á¶ÀÎÁ¤
û¼Ò³âº¸È£Ã¥ÀÓÀÚ : Á¶¾çÁ¦
Copyright © 2024 °­³²Æ÷½ºÆ®. All rights reserved.
Back to Top