TORONTO — Deposition expertise performs a key function in advancing reminiscence gadgets, and in terms of 3D NAND, stacking exposes the constraints of present filling strategies.
Lam Analysis’s just lately introduced Striker FE enhanced atomic layer deposition (ALD) platform addresses semiconductor manufacturing challenges for 3D NAND in addition to DRAM. It employs superior dielectric gapfill expertise the corporate has dubbed “ICEFill” for filling 3D NAND and DRAM buildings — in addition to logic gadgets — in rising nodes. The necessity for gapfill strategies isn’t new, stated Aaron Fellis, vp and normal supervisor of Dielectric ALD merchandise, however the conventional ones now not meet right this moment’s wants, particularly as 3D NAND is stacked greater. “They’re so tall and so they have quite a few totally different options that get etched via them to allow the mixing of various steps,” he stated. “In the end, they should get crammed again up with a dielectric materials, mostly silicon oxide.”
Legacy methods, similar to chemical vapor deposition, diffusion/furnace, and spin-on processes which might be usually used as gapfill for semiconductor manufacturing are now not viable for 3D NAND, Fellis stated, attributable to trade-offs between high quality, shrinkage, and gapfill voids. “They have a tendency to shrink and deform the precise construction that the client is constructing and designing.”
Silicon oxide continues to be a most popular materials for filling as a result of it’s steady, withstands a variety of temperatures, and displays good electrical efficiency. What’s modified is the approach for depositing it. Within the case of Lam Analysis’s Striker ICEFill, it’s a proprietary floor modification approach to realize extremely preferential bottom-up and void-free gapfill whereas retaining the movie high quality inherent to ALD, he stated. “When you use a typical ALD you get a a lot greater high quality movie as deposited, so that you don’t must cope with shrinkage.”
However even with good materials density that in flip helps good mechanical integrity inside it, stated Fellis, customary ALD can nonetheless result in gaps in some gadgets and there’s some doubt as to its extendibility. ICEFill nonetheless offers excellent high quality movie inside with out shrinkage due to its backside up strategy. “Its extendibility could be very excessive.” This implies there’s full gapfill at no matter step it’s required, he stated, whether or not it’s for mechanical strengthening of a construction or electrical efficiency. “You will have a uniform and constant materials property in a specific hole contained in the fabricated gadget.”
Deposition methods for reminiscence gadgets have their very own roadmaps which might be pushed by the development of varied reminiscence applied sciences that decide how present methods will final, stated Fellis. “Issues maintain getting taller or smaller.” Within the case of the challenges created by 3D NAND being stacked taller, Lam Analysis already had been working evolving its Striker product, he stated, “however we noticed a requirement from our buyer that we wanted to enhance the movie efficiency as they moved alongside their very own roadmaps. Stacking nonetheless drives steady innovation.”
Lam Analysis is a dominant participant for ALD expertise, stated Risto Puhakka, president of VLSIresearch, and the calls for of its expertise mirror these positioned on reminiscence. It’s all about rising density for functions, such as artificial intelligence, that require extra bits whereas retaining prices the identical, and that features gapfill capabilities because the reminiscences similar to 3D NAND are stacked greater, he stated. “The stacking turns into increasingly difficult from the manufacturing perspective, however the chip makers themselves get it just a little bit little anxious about how a lot they must spend.” Sticking with a identified materials similar to silicon oxide provides some predictability as a result of it’s nicely understood
However simply as 3D NAND stacking will finally hit limits, so will the gapfill methods and ALD expertise, added Puhakka. “It has its personal roadmap and limitations.”
Gary Hilson is a normal contributing editor with a concentrate on reminiscence and flash applied sciences for EE Times.
AI Drives Reminiscence Interconnect Evolution
New NVM Structure Might Open Up Xpoint Market
Reminiscence Makers and Foundries Step As much as Hyperscale Calls for