The past of AI chip, now with future

Drive medical treatment conflict, chemical synthesis, convict identifying, automatically wait for applied domain, the might of AI expands with each passing day. What is AI at present impracticable, future what thing may AI accomplish again? American business interrelates with systematic chip the CTO Ty Garibay of IP supplier ArterisIP published guest of a rich, explained AI is reachedAI chippreexistence this life.


Experienced the AI technology that period of first time bubble, cold winter, research restarts, current breakthrough drop is in of chip of fundamental layer AI in updating. AI can surpass the high level that reachs deepness study smoothly now after all, still remain of deliberate and test and verify.


Brief history


The term of ” of “ artificial intelligence was born 1956, be in England by 3 scientists John McCarthy, Claude Shannon And Marvin Minsky Dutt puts forward on Mao Si conference. At that time that time evening, arthur Samuel created “ machine to study the idea of ” , point to the program that can study from inside the mistake, this program can behave the person that surpasses process designing in the game such as Chinese checkers even.


The flying progress of computer technology makes investigator believes, AI can be solved very quickly. Scientists are based on computation of human brain function to whether can solve real life in research in medium problem process, created the concept of ” of “ nerve network.


1970, scientist Marvin Minsky is in ” life weekly ” in interviewing, express, 3 to 8 years inside, appear hopeful the machine that an as average as the mankind intelligence comparatives.


Go up century 80 time, AI walked out of a lab and the trend is commercialized, still lifted AI to invest mad tide. When the science and technology related AI after bubble is undone, AI returns a lab afresh again. ” of “AI cold winter comes. The personage inside course of study thinks, develop AI technology at that time too lead, cannot gain ground forever to this technology very.



1986, the father Geoffrey Hinton of nerve network and other researcher published the report of a milepost type, the report was discussed how does algorithm of retrorse transmission ” make response of deep-seated nerve network more remarkable in “ .


1989, the Yang Likun of American computer scientist that deepness learns one of 3 tycoons (Yann LeCun) as neurological as what his colleague in Bell lab can identify handwritten ZIP to pile up through education at that time and test and verify an AI theory is in the feasibility in real life.


2009, the Rajat Raina of Stanford university, Anand Madhavan and Andrew Ng published a paper, the deepness study ability that proved contemporary GPU exceeds CPU far. AI main forces appears stage a come back.




Why to invest a bound now talking about AI, element of drive of in the final analysis is computer technology maturity, huge data is obtained easily, once investigator had these resource, algorithm and solution also are not an incredible story.


But the biggest challenge of AI chip is, how be in the chip of systematic level, coordinate above resource together, and the chip of systematic level is carrier with hardware accelerator.


The design demand of AI chip is very so high, be in especially automobile industry, the standard to safety and dependability cannot be reduced.


Such company is opposite Gu Ge and tesla integrated circuit design is immature still perhaps, and although the enterprise of AI newly established such as AI Movie, Horizon Robotics has very deep attainment to machine study, but the research and development of systematic level chip that should achieve high level is very difficult also.


Photograph the accelerator of chip of deepness study AI in resembling a head to be with the buy before the car exemple, this chip is main utility is to analyse and detect the car on road, object. Every AI chip contains memorial document in order to ensure the biggest bandwidth.


Mechanism of the interconnection inside chip is detecting wider bandwidth assures when the object, and be in did not encounter an object, optimize time of consume of the sources of energy to match narrower bandwidth. And optimized method updates more advanced algorithm namely. Such AI algorithm should be updated everyday or upgrade.


If the deepness now is algorithmic banana of chip assimilate to, so nobody is willing to withhold rotted banana. Actually old algorithm is just like sodden banana in AI chip. Be opposite so for AI chip, come out time is more more sensitive than the semiconductor of other category.




Promoted the process of AI technology considerably besides deepness study and nerve network, still a lot of investigator think, if AI wants to achieve better demand, still need more methods to support AI chip.


What the design of most AI chip just is based on the scholar such as Yang Likun and Hinton at present is academic and what research and development goes out is different version, but if advance along this one orbit all the time, AI technology cannot have bigger substantial breakthrough, prep let alone replaces human thinking completely with AI.


At present AI technology still wants those who go up to build the base that transforms ” data in “ label, it cannot finish a job that is without correlation with historical experience. Nerve network already also will not have knowledge and new regulation (for example, “ up the contrary of ” is “ is down ” , child to be borne by parents wait for a fact) .


AI technology still cannot solve a problem according to doing not have the data of label now, although had not been ironed,be just like a person, also won’t touch ignition stove actively, AI still is not done however. Although pass “ label to change the data study of ” , data example also wants enough big.


AI chip fails to be behaved it seems that than the mankind more intelligence, but their study ability is very strong, future can become cleverer. The design of algorithm and chip system can progress, chip of this need AI has more advanced memory and join mechanism, and bear the weight of the hardware accelerator that deepness study data sheds.



Add a Comment

Your email address will not be published. Required fields are marked *