Flex­i­ble mi­cro­pro­ces­sor

It’s a one-bit mi­cro­pro­ces­sor with four in­struc­tions, but it could open the way to more flex­i­ble elec­tron­ics, writes Peter Sayer

PC Advisor - - CONTENTS -

Re­searchers have built a prim­i­tive mi­cro­pro­ces­sor out of a twodi­men­sional ma­te­rial sim­i­lar to graphene, the flex­i­ble con­duc­tive won­der ma­te­rial that some be­lieve will rev­o­lu­tionise the de­sign and man­u­fac­ture of bat­ter­ies, sen­sors and chips.

With only 115 tran­sis­tors, their pro­ces­sor isn’t go­ing to top any bench­mark rank­ings, but it’s “a first step to­wards the devel­op­ment of mi­cro­pro­ces­sors based on 2D semi­con­duc­tors,” the re­searchers at Vi­enna Univer­sity of Tech­nol­ogy said in a pa­per pub­lished in the jour­nal Na­ture.

Two-di­men­sional ma­te­ri­als have the ben­e­fit of flex­i­bil­ity, mean­ing that they can be in­cor­po­rated more eas­ily into wear­able de­vices or con­nected sen­sors, and po­ten­tially mak­ing them less break­able: Pic­ture a smart­phone that bends rather than breaks if you drop it.

To­day’s semi­con­duc­tors and screens are al­ready pretty thin, but they still rely on the three-di­men­sional phys­i­cal prop­er­ties of the ma­te­ri­als they’re made from in or­der to func­tion. Bend a sil­i­con wafer and it will crack. But 2D ma­te­ri­als like graphene or the tran­si­tion-metal dichalco­genide (TMD) used by the Vi­enna re­searchers, are truly two-di­men­sional, made with crys­tals just one layer of atoms or mol­e­cules thick, al­low­ing them to flex.

TMDs are com­pounds com­posed of a tran­si­tion metal such as molyb­de­num or tung­sten and a chalco­gen (typ­i­cally sul­phur, se­le­nium or tel­lurium, al­though oxy­gen is also a chalco­gen). Like graphene, they form into lay­ers, but un­like graphene which con­ducts elec­tric­ity like a metal, they are semi­con­duc­tors, which is great news for flex­i­ble chip de­sign­ers.

Ste­fan Wachter, Dmitry Polyushkin and Thomas Mueller of the In­sti­tute of Pho­ton­ics, work­ing with Ole Bethge of the In­sti­tute of Solid State Elec­tron­ics in Vi­enna, de­cided to use molyb­de­num disul­fide to build their mi­cro­pro­ces­sor.

They de­posited two mol­e­cule-thick lay­ers of it on a sil­i­con sub­strate, etched with their cir­cuit de­sign and sep­a­rated by a layer of alu­minium ox­ide.

“The sub­strate ful­fils no other func­tion than act­ing as a car­rier medium and could thus be re­placed by glass or any other ma­te­rial, in­clud­ing flex­i­ble sub­strates,” they wrote.

Re­cent In­tel mi­cro­pro­ces­sors act on data in 64-bit ‘words’, can un­der­stand hun­dreds or even thou­sands of dif­fer­ent in­struc­tions, de­pend­ing on how you count them, and con­tain hun­dreds of mil­lions of tran­sis­tors.

In con­trast, the mi­cro­pro­ces­sor built by the re­searchers is only ca­pa­ble of act­ing on data one bit at a time, us­ing a set of just four in­struc­tions (NOP, LDA, AND and OR), and the cir­cuit fea­tures used to build it are of the or­der of two mi­crom­e­ters across, 100 times larger than those found in the lat­est In­tel and ARM pro­ces­sors. With more work, though, the mi­cro­pro­ces­sor’s com­plex­ity could be in­creased and its size re­duced, the re­searchers said. They de­lib­er­ately chose an overly large fea­ture size for their man­u­fac­tur­ing process to re­duce the ef­fects of holes, cracks and con­tam­i­na­tion in the molyb­de­num disul­fide film and to make it eas­ier to in­spect the re­sults with an op­ti­cal mi­cro­scope.

“We do not see any road­blocks that could pre­vent the scal­ing of our 1-bit de­sign to multi-bit data,” they said, and only the chal­lenge of low­er­ing con­tact re­sis­tance stands in the way of sub­mi­crom­e­ter man­u­fac­tur­ing.

That’s not to say it will be easy: al­though the man­u­fac­tur­ing yield for sub­units was high, with around 80 per­cent of the arith­metic-logic units fully func­tional, their non-fault tol­er­ant de­sign meant only a few per­cent of fin­ished de­vices worked prop­erly.

Com­mer­cial mi­cro­pro­ces­sor man­u­fac­tur­ers deal with yield prob­lems by mak­ing their chip de­signs mod­u­lar, and test­ing them at a va­ri­ety of speeds. Pro­ces­sors that work at higher speed fetch higher prices, while faulty sub­com­po­nents can be per­ma­nently dis­abled and the re­sult­ing chips, oth­er­wise fully func­tional, sold as lower-spec­i­fi­ca­tion mod­els.

It’s taken 46 years for In­tel to get from the 4004, a four-bit cen­tral pro­ces­sor with 46 in­struc­tions, to the lat­est in­car­na­tion of the x86 ar­chi­tec­ture, Kaby Lake: with all that the in­dus­try has learned about mi­cro­man­u­fac­tur­ing since then, progress with flex­i­ble semi­con­duc­tors may be a lit­tle faster.

Newspapers in English

Newspapers from UK

© PressReader. All rights reserved.