THE SMARTER CHIP

SMART­PHONES TO GET SMARTER THAN EVER WITH DED­I­CATED HARD­WARE FOR AI COM­PUT­ING IN CHIPSETS.

Gadgets and Gizmos (India) - - CONTENTS - By NIDHI SIN­GAL @nid­hisin­gal

Smart­phones now come equipped with ded­i­cated AI hard­ware.

Imag­ine emo­jis that you can con­trol with your face – the adorable panda or dog emoji that looks and talks like you. One of the high­lights of the much-awaited iPhone X is the An­i­moji fea­ture. Us­ing the TrueDepth cam­era that analy­ses more than 50 fa­cial mus­cle move­ments, users will be able to lend their face and ex­pres­sions to a dozen emo­jis, edit and send them out in real time. Pow­er­ing this fancy new fea­ture on the Ap­ple iPhone X is the A11 Bionic chip with neu­ral en­gine that can recog­nise peo­ple, places and ob­jects.

A neu­ral en­gine, ex­plains Ap­ple, is a hard­ware that’s “pur­pose-built for ma­chine learn­ing, a type of ar­ti­fi­cial in­tel­li­gence that en­ables com­put­ers to learn from ob­ser­va­tion. It’s ca­pa­ble of in­cred­i­bly fast com­pu­ta­tions needed by neu­ral net­works while also be­ing in­cred­i­bly ef­fi­cient.”

Hav­ing fo­cused on mul­ti­ple cores to boost per­for­mance in smart­phones, chip man­u­fac­tur­ers are now adding ded­i­cated hard­ware to ad­dress the need for ma­chine learn­ing to han­dle tasks such as real-time voice pro­cess­ing and im­age recog­ni­tion swiftly. Th­ese new com­po­nents im­prove the per­for­mance and ef­fi­ciency of tasks as­so­ci­ated with AI as­sis­tants. Although ex­ist­ing processors can be put to use, they slow down the de­vice and drain the bat­tery much faster.

With smart­phones in­clud­ing chipsets with ded­i­cated hard­ware for AI, users can ex­pect a sig­nif­i­cantly bet­ter ex­pe­ri­ence. Smart­phones will be able to lis­ten, read and un­der­stand how users think, and pro­vide more rel­e­vant in­for­ma­tion when needed in real time. For in­stance, based on the time of day, an AI en­gine can sug­gest switch­ing to low-light mode to avoid eye strain.

The neu­ral en­gine in iPhone X’s A11 Bionic chip helps un­lock­ing the phone us­ing FaceID, ir­re­spec­tive of a change in the user’s ap­pear­ance. It is also equipped to carry out in­tu­itive tasks such as keep­ing the screen lit when you’re read­ing, and low­er­ing the vol­ume of an alarm or ringer when you are look­ing at the de­vice with eyes wide open. There is a ‘Hey Siri’ de­tec­tor that uses a deep neu­ral net­work to ac­ti­vate Ap­ple’s voice as­sis­tant, Siri, with­out the user hav­ing to press a but­ton. Hard­ware, soft­ware, and In­ter­net ser­vices work seam­lessly to­gether to pro­vide this hands-free ex­pe­ri­ence.

Huawei’s Kirin 970 chipset has a ded­i­cated neu­ral net­work pro­cess­ing unit (NPU) that has ad­vanced ob­jec­trecog­ni­tion func­tion. It helps iden­tify ob­jects in real time to ad­just metrics for cap­tur­ing bet­ter pho­tos or of­fer real-time trans­la­tions in aug­mented re­al­ity mode. “The aim is to de­liver this in the most ef­fi­cient way with on-de­vice AI pro­cess­ing in or­der to en­sure an ex­pe­ri­ence with less la­tency, less power con­sump­tion, no net­work de­pen­dence and more pri­vacy. This new gen­er­a­tion of smart­phones will trans­form hu­man-toma­chine in­ter­ac­tions,” ex­plains a Huawei spokesper­son. Huawei Mate 10 and Mate 10 Pro are the first two smart­phones launched with the AIpow­ered Kirin 970 chip.

Qual­comm has adopted a dif­fer­ent ap­proach. It has cre­ated a soft­ware de­vel­op­ment kit (SDK) for de­vel­op­ers to power im­mer­sive and en­gag­ing user ex­pe­ri­ences with ma­chine learn­ing on its mo­bile plat­forms us­ing the Snap­dragon Neu­ral Pro­cess­ing En­gine (NPE). This means de­vel­op­ers can take ad­van­tage of user ex­pe­ri­ences across fil­ters (aug­mented re­al­ity), scene de­tec­tion, fa­cial recog­ni­tion, nat­u­ral lan­guage un­der­stand­ing, ob­ject track­ing and avoid­ance, ges­tur­ing and text recog­ni­tion, to name a few, and op­ti­mise their apps to run AI ef­fi­ciently.

In­stead of fo­cus­ing only on the flag­ship de­vices, Qual­comm’s Snap­dragon NPE SDK sup­ports mul­ti­ple chips in­clud­ing ones that power mid-level smart­phones. “The pri­mary ben­e­fit of on-de­vice AI is a more seam­less, im­mer­sive, re­al­is­tic and real-time ex­pe­ri­ence,” says a Qual­comm spokesper­son.

Google’s lat­est – Pixel 2 and Pixel 2 XL – have a spe­cial-pur­pose chip for AI-re­lated tasks and im­age pro­cess­ing. The Pixel Vis­ual Core co­pro­ces­sor is be­lieved to im­prove speed and bat­tery life when shoot­ing pho­tos with Google’s HDR+ tech­nol­ogy. Sam­sung isn’t far be­hind. Re­cently, the com­pany in­vested in Chi­nese AI firm DeePhi Tech. Sam­sung’s next Galaxy S9 flag­ship, to be launched in 2018, is ru­moured to be pow­ered by the by Exynospow­ered AI chip.

By in­te­grat­ing smart de­vices with the cloud and big data, on-de­vice AI will con­tinue to make ad­vance­ments in the realm of se­cu­rity, pro­cess­ing, and low power con­sump­tion. Cur­rently, a lot of data is be­ing sent to the cloud for au­then­ti­ca­tion, but with AI-em­bed­ded chipsets that en­able real-time pro­cess­ing, there are lesser chances of data be­ing leaked or hacked. Smart­phones are slated to be­come smarter and more sen­si­tive in the days ahead.

Ap­ple A11 Bionic chip's neu­ral en­gine han­dles AI-based tasks, pow­ers FaceID and An­i­moji, among other fea­tures

Huawei Kirin 970 chipset has a ded­i­cated neu­ral net­work pro­cess­ing unit that has ad­vanced ob­ject-recog­ni­tion func­tion

Newspapers in English

Newspapers from India

© PressReader. All rights reserved.