`Proc. of SPIE Vol. 9779, 97790N · © 2016 SPIE · CCC code: 0277-786X/16/$18 · doi: 10.1117/12.2218696
`Page 1 of 12
`Proc. of SPIE Vol. 9779 97790N-1
`
`
`
`3D-ICs created using oblique processing
`D. Bruce Burckel
`Sandia National Laboratories, P.O. Box 5800, Albuquerque, NM, USA 87106
`
`ABSTRACT
`This paper demonstrates that another class of three-dimensional integrated circuits (3D-ICs) exists, distinct from
`through silicon via centric and monolithic 3D-ICs. Furthermore, it is possible to create devices that are 3D at the
`device level (i.e. with active channels oriented in each of the three coordinate axes), by performing standard CMOS
`fabrication operations at an angle with respect to the wafer surface into high aspect ratio silicon substrates using
`membrane projection lithography (MPL). MPL requires only minimal fixturing changes to standard CMOS
`equipment, and no change to current state-of-the-art lithography. Eliminating the constraint of 2D planar device
`architecture enables a wide range of new interconnect topologies which could help reduce interconnect
`resistance/capacitance, and potentially improve performance.
`
`Keywords: 3D-ICs, interconnects, oblique processing
`
`
`1. INTRODUCTION
`Moore’s law 1, an observation that the cost per transistor decreases as transistor density increases, following a
`roughly 2 year doubling period, has dominated the landscape of semiconductor research since it was first proposed
`in 1965. For much of this run, Moore’s law was supported by Dennard scaling 2 which posits that processing
`performance per Watt increases with decreasing device dimension, which also possesses a similar doubling period.
`Thus the reduced cost per transistor was accompanied by higher performance per Watt, a win-win proposition which
`served as a self-sustaining feedback mechanism, responsible for today’s massive semiconductor and personal
`electronics industries among other epochal changes. This coincidental scaling of cost, performance and power
`consumption allows these trends to be conveniently plotted on familiar log-linear graphs showing the breathtaking
`ascent and, more recently, inevitable stall of these curves as the semiconductor industry has eclipsed the 28 nm
`processing node.
`
`Transistors with smaller device dimensions are subject to a variety of deleterious effects such as drain induced
`barrier lowering (DIBL), gate leakage, subthreshold leakage, etc. responsible for the end of Dennard scaling. Even
`though subsequent lithography nodes add higher transistor density, the failure of these new device designs to scale
`power requirements means that not all of these transistors can be concurrently active for the same power budget,
`leading to the notion of dark silicon, a problem which scales non-linearly with shrinking dimensions. In addition to
`these deterministic effects, small transistors also incur increased variance of the statistical distribution of device
`performance. Increasing variability in performance from device to device complicates the already enormous task of
`designing circuits and multi-circuit modules. 3
`
`Pursuit of Moore’s law has required more than just smaller transistors. Adoption of increasingly complicated
`interconnection strategies has also been necessary in order to allow the dense sea of transistors to communicate with
`one another. For some time it has been recognized that gate delay and transistor capacitance is only a fraction of the
`overall delay in switching speed.4 Interconnect capacitance has dominated gate delay, forcing the adoption of copper
`damascene and pursuit of low-k dielectrics. The culmination of this trend is the inclusion of air-gaps to further
`reduce capacitance by filling the space between interconnects with a material with the lowest possible dielectric
`constant at the cost of manufacturing complexity, yield, and potential reliability concerns.
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Page 2 of 12
`
`Proc. of SPIE Vol. 9779 97790N-2
`
`
`
`Modern integrated circuits are highly optimized to maximize performance per unit area. Shrinking device
`dimensions and areal scaling are achieved through process node improvements, fueling much of this optimization,
`however, design process technology co-optimization, the practice of using process-aware design of circuit layout has
`become increasingly important5. Adoption of Manhattan layout geometry and self-aligned double/quadruple
`patterning (SADP, SAQP) are examples where full 2-dimensional interconnect spatial freedom are sacrificed in
`order to enable advanced lithography techniques. Optimized module designs are constructed with the optimized
`transistors. In order to connect these modules together, sophisticated place and route algorithms are used to construct
`larger computational blocks.
`
`These increasingly disruptive trends have forced the semiconductor industry to adopt new materials, process
`techniques and device design concepts, with even more drastic changes required for future process nodes.6 Adoption
`of a 3D integrated circuit topology is seen by many in the industry as a viable approach to continue density scaling.
`The shorter average interconnect lengths of 3D-ICs 7 reduce both the resistance and capacitance of the line, and
`hence reduce ohmic power dissipation and RC time delay. While just on the verge of being adopted as a high
`volume manufacturing (HVM) solution, the concept and advantages of 3D-ICs have been identified since at least the
`1980’s 8. In its current incarnation, 3D-IC architectures are divided into two distinct types: 1) 3D-ICs created by
`stacking planar 2D chips 9, achieving interconnection in the vertical direction using through silicon vias (TSV); and
`2) monolithic 3D-ICs where epitaxial regrowth is performed on the wafer after fabrication of the first layer of
`devices, yielding a second layer of single crystal silicon for a second layer of silicon devices 10. Both of these
`approaches yield ICs with current flow vertical to the wafer surface, and, in that sense are 3-dimensional, however
`the transistors forming each integrated circuit all have their active regions oriented parallel to the wafer surface.
`
`The purpose of this paper is to point out that another class of 3D-ICs exists with transistors oriented along each of
`the coordinate axes. Distinct from both TSV-3D-ICs and monolithic 3D-ICs, these take full advantage of the third-
`dimension at the device and module level, to affect increases in trace width and trace separation, reducing
`interconnect resistance and capacitance, while capturing the inherent reduction in average interconnect length that
`comes with 3D interconnection. Each of these help to improve the interconnect delays and power dissipation which
`threaten to make further lithography node scaling ineffectual. Furthermore, we advance an oblique processing
`approach, membrane projection lithography (MPL), as a fabrication method capable of fabricating this new class of
`3D-ICs, requiring only minimal fixturing changes to current state-of-the-art semiconductor fabrication equipment.
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Page 3 of 12
`Proc. of SPIE Vol. 9779 97790N-3
`
`
`
`2. ADVANTAGES OF DEVICE-LEVEL 3D-ICS
`
`
`
`Figure 1. (A) Planar logic module W x 3W in size. Interconnect from point 1 to point 2 is 3W micrometers long; (B)
`Folding the two edges of the planar module allows for module level three dimensionality; (C) Same module folded into 90°
`segments. The distance from point 1 to point 2 is only W micrometers long.
`
`Even in two-dimensional topologies, layout, placement and routing are all enormously complex endeavors. In the
`discussion that follows, no attempt has been made to generate optimal designs or to consider the myriad co-
`dependent constraints that exist in an actual circuit design, but rather to advance 3-D specific design possibilities
`that cannot be adopted by a two-dimensional approach, or for that matter, 3D-ICs in either a TSV or monolithic
`approach.
`
`Consider the planar multi-module layout in Fig. 1(A). This layout block consists of three W-by-W micrometer
`sections with many transistors. The areal footprint of this block is 3W 2. Furthermore, an interconnect is required
`from points 1-2 (noted by the yellow stars in the figure). Given the dimensions of the block, this interconnect will be
`3W micrometers in length, and the trace will be stood off from the metal layers beneath it by the height of the
`underlying metal layer, typically in the deep sub-micrometer regime. Without addressing how this is to be done for
`the moment, assume that the two end sections of the block are folded up (Fig. 1(B)) so that they form right angles
`with the middle section as shown in Fig. 1(C). The device layer of silicon for the two edge sections has a thickness
`Δ, and the current flow for the transistors in the two edge sections is contained in the xz-plane rather than the xy-
`plane. The block in Fig. 1(C) has an areal footprint of W2+2ΔW, so that for Δ << W, the configuration in Fig. 1 (C)
`has nearly a factor of 3 increase in areal transistor density versus the planar case in Fig. 1(A).
`
`The interconnect from points 1-2, is now shortened from 3W to W, a factor of 3 reduction, while the separation of
`the trace connecting these two points with the underlying metal layers becomes W, significantly greater than in the
`planar case. Shortening the interconnect length reduces ohmic loss and distributed capacitance. Furthermore, the
`increased separation reduces capacitance and cross talk. In a planar geometry, the separation between traces is fixed
`by the metal thickness. To lower the capacitance between traces in a 2-D geometry, the only knob to tweak when the
`distance between traces is fixed is to pursue inter-layer dielectrics (ILD) with smaller dielectric constants. A
`meaningful reduction in dielectric constant over dense ILDs is only achieved by incorporation of porosity, which
`necessarily impacts process robustness. Given that the relative dielectric constant of air is 1 and SiO2 is 4, the
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`t
`
`t T
`
`d
`
`(A)
`
`(B)
`
`Page 4 of 12
`Proc. of SPIE Vol. 9779 97790N-4
`
`
`
`maximum achievable reduction in capacitance through material changes is 4. In reality, SiO2 has already been
`replaced by materials with εr ~2, so that there is much less room for further improvement through material selection.
`
`Figure 2. (A) Two horizontally stacked copper interconnect traces with dimensions W x L x t, separated by a distance d; (B)
`Two vertically stacked copper interconnect traces with dimensions W x L x t, separated by a distance d, with significantly
`reduced capacitance.
`
`
`
`Exploring the advantages of 3D topology even further, consider Fig. 2. In Fig 2(A), two parallel traces with
`thickness, t, are oriented horizontally and stacked vertically. Treating these two traces as a parallel plate capacitor,
`the capacitance between the two traces is given by
`
`
`
`
`
`
`
`
`
`
`
`
`
`(1)
`
`(cid:1829)(cid:2869),(cid:2870)=(cid:2013)(cid:3002)(cid:3117),(cid:3118)(cid:3031)
`
`where A is the cross sectional area of the plates (given by A1 = W x L), d is the separation of the plates and ε is the
`dielectric constant of the material separating the traces. In Fig 2(B), the same traces are oriented vertically and
`stacked vertically, separated by the same distance, d. In this case the cross sectional area between the two traces is
`given by A2 =- t x L, so that for t << W, the capacitance C2 << C1. Since the physical dimensions of the traces in
`both cases are identical, they both possess equivalent current carrying capacity and resistivity. Furthermore, consider
`the areal footprint of the traces as measured in the xy-plane. The traces in Fig 2(A) have a footprint of W x L
`whereas the traces in Fig. 2(B) have a footprint of (Δ + t ) x L, which, if (D + t) < W, has the same current capacity
`and resistance, smaller capacitance and a smaller footprint.
`
`As a further example of the possible leverage a fully 3D approach has, consider Fig. 3(A), where a trace with a
`thickness t is deposited over a silicon feature with height h and width Δ. The resulting trace has a cross-sectional
`area proportional to 2 x h, while occupying an areal footprint of just (Δ + 2t). Another advantage of a “spine”
`interconnect such as this is in addition to the large cross-sectional area is that it serves as a common interconnect for
`devices on the left and right side of the spine simultaneously. It is also possible to create coaxial or core-shell
`interconnects like those shown in Fig. 3(B), where significant shielding is gained. While modern interconnects bear
`little resemblance to the parallel plate traces of Fig. 2(A), the principles and potential advantages remain.
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Inner
`Conductor
`
`Dielectric
`Separator
`4_S
`
`Outer
`Conductor
`
`(B)
`
`(A)
`
`(B)
`
`Page 5 of 12
`Proc. of SPIE Vol. 9779 97790N-5
`
`
`
`Figure 3. (A) Vertical “spine” interconnect with surface area 2 x h; (B) Coaxial vertical spine interconnect for propagating
`“shielded” signals.
`
`
`
`
`
`3. DEVICE LEVEL 3D-IC GEOMETRY
`
`
`
`Figure 4. (A) One possible etched, high aspect ratio (HAR) silicon matrix for a “non-folded” fabrication approach to device
`level 3D-ICs; (B) HAR silicon matrix including orthogonal interconnect spines.
`
`From Section 2, we saw that the folded, 3D version of the logic block has a ~3X increase in areal transistor density,
`which is attractive. However, aside from the complete lack of manufacturability of such a folding scheme, this
`approach has at least one more fatal flaw: neighboring modules cannot be densely packed together so that the chip
`level density remains the same. This can be remedied in a highly manufacturable way, however, as 3D-ICs can be
`created in a dense array of 3D modules by fabricating the transistors in high aspect ratio (HAR) machined silicon.
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Hitching
`Post
`Vias
`
`(B)
`
`Copper
`Power
`Rails
`
`(A)
`
`Page 6 of 12
`
`Proc. of SPIE Vol. 9779 97790N-6
`
`
`
`Fig. 4(A) shows two back-to-back 3D modules in just one of several possible HAR 3D motifs. In this case, each
`module has a device silicon thickness of Δ, while adjacent modules are separated by a gap of distance Γ. The length,
`L, of the modules is a free parameter, while the height, h, is constrained by the resultant aspect ratio (AR) of either
`the vertical device layer or the gap, depending on their dimensions. Etched aspect ratios of 10:1 are fairly routine. In
`principle long modules with L >> W can be fabricated, however it might be advantageous to include orthogonally
`positioned spines with a separation Γ2, from the vertical device module fin as shown in Fig. 4(B).
`
`Figure 5. (A) 3D logic block combined with spine interconnects for global distribution of power; (B) Examples of
`“hitching post” vias, to allow for inter-module interconnects and vertical signal routing.
`
`
`
`
`
`Combining the HAR structure with 3D interconnect motifs allows construction of larger functional blocks. Fig. 5(A)
`shows a single module with large surface area, spine interconnects. Whereas global signals such as the power rails
`and clock signals are typically distributed at the highest metal level and then multiplexed down to the individual
`transistors, in this instance, these spines can be used to route global signals down at the silicon level (L0 metal?).
`Inclusion of “hitching post” vias in the Γ1 gap, allows for short interconnection between adjacent 3D modules,
`further reducing interconnect length (Fig. 5(B)). These vias can be formed in the same step as the power rail
`formation as long as suitable isolation is introduced. By etching through the device and isolation layers on both sides
`of the via, interconnection between the two faces is achieved.
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Cross -Spine
`Interconnect
`
`(A)
`
`(B)
`
`Page 7 of 12
`Proc. of SPIE Vol. 9779 97790N-7
`
`
`
`
`
`Figure 6. (A) Larger chip level construct showing interleaved spine interconnects; (B) Backfilled and CMP’d structure with
`higher level back end of the line (BEOL) metallization for cross spine and regional interconnects.
`
`Fig. 6 contains the vision of larger chip level constructs built up from many multi-modules. In Fig. 6(A),
`neighboring multi-module blocks are separated by spine interconnections with global interconnects. The double-
`sided spines allow the global signals to be interleaved. Fig. 6(B) shows the same blocks backfilled and planarized
`via CMP. At this point standard back end of line (BEOL) processing can be used to build up intermediate length
`interconnects and higher layer global interconnects.
`
`4. MEMBRANE PROJECTION LITHOGRAPHY
`With the geometry and some interconnect strategies established, the only remaining question is “How do we create
`transistors on the vertical faces of silicon?” Fabrication of CMOS devices requires “blanket” process steps such as
`oxidations and CVD/ALD depositions, as well as patterned, directional steps such as ion implantation, dry etching
`and metal deposition. Since most blanket steps such as oxidation, and CVD /ALD deposition can be performed
`conformally, demonstration of oblique versions of ion implantation, dry etching and metal deposition enables
`fabrication of MOSFETs in high aspect ratio silicon topography. The key to achieving oblique processing is
`membrane projection lithography (MPL), a technique which creates suspended inorganic membranes patterned with
`the desired pattern over 3D topography etched in single crystal silicon using standard CMOS equipment 11-13.
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Silicon
`
`Etch
`Oxide
`Topography Backfill
`
`CMP
`Flat
`
`Deposit
`AIN
`
`Pattern
`Membrane
`
`Evacuate
`Backfill
`
`Ready
`For Process
`
`(A)
`
`(B)
`
`SiOz
`
`(C).'' ..
`.ó
`
`O'.' '
`
`Ion
`Implant
`
`Metal
`Deposition
`
`Dry
`Etch
`
`Page 8 of 12
`Proc. of SPIE Vol. 9779 97790N-8
`
`
`
`
`
`Figure 7. Schematic diagram of the membrane projection lithography process flow.
`
`Fig. 7 shows a schematic sequence of the MPL process for metal deposition. Starting from planar silicon (A), the
`desired topography is etched (B), backfilled with CVD oxide (C) and chemically mechanically polished (CMP) flat.
`An aluminum nitride film is deposited (E), patterned using standard lithography and etched (F). The backfill oxide is
`then evacuated using hydrofluoric acid (G). At this point, the AlN film exists as a patterned membrane suspended
`over the HAR silicon, serving as a stencil for patterning the underlying silicon through ion implantation, metal
`deposition or dry etching. After processing, the membrane is removed using a version of the standard SC1 clean
`(H2O:NH4OH: H2O2, 5:1:1 at 70 °C)
`
`Both ion implantation and metal deposition (sputtering and e-beam evaporation) are highly directional processes so
`that positioning the substrate at an angle with respect to the incoming flux results in either implantation or
`deposition through the membrane onto the vertical sidewall. Dry etching is different than either implantation or
`metallization due to the formation of a plasma used to create and accelerate the etchant species toward the etch
`platen. During etching, a plasma sheath conforms to the substrate, accelerating the ions normal to the substrate.
`Using a Faraday cage, the direction of the ion acceleration can be altered so that oblique etching will occur14-16.
`
`Fig. 8 contains proof-of-concept demonstrations of the three directional steps of ion-implantation, metal deposition
`and dry etching. Fig. 8(A) contains the results of a process simulation (Athena) showing patterned implant through a
`1000 Å thick nitride membrane. While the composition and thickness of the membrane and implant dose have not
`been optimized, the ability of a thin membrane to successfully define a high-dose implant region on the vertical
`sidewall is established. In Fig. 8(B) and (C), SEM images of deposited metallic (B) and etched features (C) are
`shown. The structures were all fabricated on 150 mm wafers, using 248 nm optical lithography. Although the HAR
`silicon patterning in these SEMs is sub-optimal, it is apparent that high fidelity patterns can be produced using this
`method.
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Implanted Features
`
`Implant
`Direction
`
`1
`
`Deposited Features
`
`MI MI
`w..11
`
`Etched Features
`
`IiiI,OlI.IIIi2
`
`V 11.0mm x25.0k SE(U) 17J172Ò14 1037
`
`2.00um
`
`(A)
`
`(B)
`
`(C)
`
`To p
`
`Wire mesh to
`allow ions to penetrate
`
`Page 9 of 12
`Proc. of SPIE Vol. 9779 97790N-9
`
`
`
`Figure 8. (A) Athena process simulation demonstrating that a 1000Å thick nitride membrane is capable of defining a
`patterned region on the vertical sidewall; (B) Tilted, cross-section SEM image of vertically oriented metal depositions on
`the silicon sidewall; (C) Tilted cross-section SEM image of etched features on a vertically oriented silicon sidewall.
`
`
`
`
`
`5. PROSPECTS FOR INTEGRATION TO HVM FABRICATION
`The MPL process demonstrated uses CMOS compatible materials and standard semiconductor processes and
`equipment. Lithographic patterning of the membrane occurs on a CMP-flat surface, compatible with any type of
`lithography including SOA high NA immersion steppers. Oblique ion implantation is used for halo implants and is
`already in HVM. For metal deposition, the only change required is a fixture for positioning the wafer at an angle to
`the source as well as methods for homogenizing the metal flux, as the wafers cannot be rotated during deposition.
`Additionally the metal process is inherently a “lift-off” process, which has been replaced in the industry by either
`blanket deposit/etch or damascene patterning. BEOL processing of interconnects can still be performed using
`standard copper damascene. The introduction of a Faraday cage into the etch chamber necessary for patterning
`sidewalls can be handled seamlessly by creating a wafer clamp ring with individual die-level Faraday cages (Fig. 9).
`Again, only a minor fixturing change, which adapts current capital equipment into a tool-set capable of oblique
`processing.
`
`Figure 9. Etch chamber clamp ring with die-level Faraday cages oriented at a 45° angle with respect to the wafer surface
`for etching vertical sidewalls.
`
`
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Page 10 of 12
`Proc. of SPIE Vol. 9779 97790N-10
`
`
`
`
`
`Patterning using the MPL approach requires that steps (C) – (G) in Fig. 7 are repeated each time a new pattern must
`be transferred. Fortunately, the industry has already embraced multiple patterning steps in order to enable the
`SADP/SAQP approach, so this is perhaps not as onerous a requirement as it might seem, and is only necessary for
`front end processing.
`
`We have yet to rigorously establish the resolution limits of MPL, but we have anecdotal evidence that < 50 nm
`isolated features are resolvable. Dense line/space patterns may be more problematic. Clearly, thinner membranes
`allow for finer features at the expense of structural integrity for the membrane. Another concern is the need for
`wafer alignment in the tool. Whereas alignment and overlay of the optical lithography to pattern the membrane
`remain unchanged from a planar process, the projection of that pattern by the etch, deposition or implant operation is
`based on the trajectory set by both the in-plane and out of plane rotation of the wafer with respect to the source, and
`must be controlled to a tight tolerance.
`
`While fabrication in the “folded” space of the HAR silicon matrix increases areal transistor density by nearly a
`factor of 3, the need for gaps, and device silicon thickness as well as (optional) interconnect spines cut in to this
`gain. Additionally, the prospects of creating a finFET transistor on the sidewalls are remote, so transistor designs
`from earlier than the 45 nm node are probably required. Finally, adopting this 3D-IC fabrication approach does not
`preclude the use of TSVs for chip/wafer stacking, or monolithic regrowth for subsequent layers of 3D transistors,
`should either approach be adopted by industry.
`
`The hope is that the reduction in interconnect resistance and capacitance won by adopting this 3D approach yields
`net advantages over higher density planar designs fabricated using the standard approach in a planar topology. While
`the story arc of Moore’s law would tend to disagree with this, some cracks are beginning to form in the prevailing
`narrative, “from at least the 45 nm node, we can create smaller logic blocks using gate pitches larger than nominal,
`owing to the combined effects of parasitics, strain, and lithography limitations” 3.
`
`6. CONCLUSIONS
`The challenges outlined in the previous section are substantial. For almost the entire history of the IC industry, such
`a re-imagining of the fabrication of transistors at the device level would have been summarily dismissed. But these
`are interesting times. Industry conferences are full of sessions considering materials to replace silicon (carbon
`nanotubes, graphene, MoS2, etc.), steeper sub-threshold devices (tunnelFETs), phase change materials (MEMristors)
`and non-von Neuman computing solutions (the last three of which usually also require non-silicon starting material).
`Such massive departures from silicon based, charge control, CMOS fabricated with optical lithography represent a
`long horizon endeavor, with many years (decades?) of research. In that light, the fact that we can generate such a
`specific, if incomplete, list of challenges to 3D-ICs fabricated with MPL could be seen as an endorsement – the
`present approach leverages 60+ years of research into n-type and p-type contacts to silicon, gate stack engineering,
`drain and source engineering, 450 mm starting material, CMP, design, layout and placement and routing. We are
`able to judge the MPL approach so critically because it exists in a space we are familiar with; we have the callouses
`and scar tissue as evidence. Some of these tried and true notions may have to be discarded or adapted, but at least
`the issues are known. Adopting a non-silicon based approach abandons much of this hard-won insight, and makes it
`difficult to assess the first order issues with a new technology, let alone the show-stopping, devil’s in the details
`issues which are more than enough to engulf and quash a seemingly promising direction.
`
`The seriousness with which these alternatives are being considered speaks to the enormity of the task to extend
`Moore’s law scaling of computing performance, if not transistor density. An alternative to continuing to
`simultaneously address cost, performance and power is to split the application space into segments which are highly
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Page 11 of 12
`Proc. of SPIE Vol. 9779 97790N-11
`
`
`
`sensitive to improvements in one axis, while being tolerant of sub-optimality in the other two. Perhaps ever-
`increasing density through device scaling was a guide star followed one or two process nodes too far for some of
`these applications. In contemplating such a fractured application space, it is possible that the 3D-IC approach
`advanced here could provide an acceptable balance between process complexity and device performance in
`applications where the density of a few lithography nodes ago, combined with reduced interconnect capacitance and
`resistance yield a superior product.
`
`
`
`Acknowledgements
`
`Supported by the Laboratory Directed Research and Development program at Sandia National Laboratories, a multi-
`program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin
`Corporation, for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-
`AC04-94AL85000.
`
`References
`
`[1] Moore, G., “Cramming more components onto integrated circuits,” Electronics Magazine, 38, 114-117 (1965).
`
`[2] Dennard, R. H., Gaensslen, F. H., Yu, H.-N., Rideout, V. L., Bassous, E. and LeBlanc, A. R., “Design of ion-
`implanted MOSFET’s with very small physical dimensions,” Proceedings of the IEEE, 87, 668-678 (1999 reprint of
`the original 1974 paper .)
`
`[3] Yeric, G., “Moore’s Law at 50: Are we planning for retirement?” IEDM Proceedings, 1.1, (2015).
`
`[4] Buchanan, K., “The evolution of interconnect technology for silicon integrated circuitry,” GaAs MANTECH
`Conference, (2002).
`
`[5] Liebmann, L., Pileggi, L., Hibbeler, J., Rovner, V., Jhaveri, T., and Northrop, G., “Simplify to survive,
`prescriptive layouts ensure profitable scaling to 32 nm and beyond,” Proc. SPIE 7275, Design for Manufacturability
`through Design-Process Integration III, 72750A (2009).
`
`[6] International Technology Roadmap for Semiconductors:Design, 2011 Edition.
`
`[7] Stroobandt, D., “Recent advances in system-level interconnect prediction,” IEEE Cir. and Sys., 11, 4-19 (2000).
`
`[8] Akasaka, Y., “Three-dimensional IC trends,” Proc. of the IEEE, 74, 1703-1714 (1986).
`
`[9] Topol, A. W., La Tulpe Jr., D. C., Shi, L., Frank, D. J., Bernstein, K., Steen, S. E., Kumar, A., Singco, G. U.,
`Young, A. M., Guarini, K. W., and Ieong, M., “Three-dimensional integrated Circuits,” IBM J. Res. & Dev., 50,
`491-506 (2006).
`
`[10] Wong, S., El-Gamal, A., Griffin, P., Nishi, Y., Pease, F. and Plummer, J., “Monolithic 3D integrated
`circuits,” Proc. Of Technical Papers, International Symposium on VLSI Technology, Systems and Applications,
`IEEE, April 2007.
`
`[11] Burckel, D. B., Wendt, J. R., Ten Eyck, G. A., Ellis, A. R., Brener, I. and Sinclair, M. B., “Fabrication of 3D
`metamaterial resonators using self-aligned membrane projection lithography,” Adv. Mater., 22, 3171-3175 (2010).
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx
`
`
`
`Page 12 of 12
`Proc. of SPIE Vol. 9779 97790N-12
`
`
`
`[12] Burckel, D. B., Wendt, J. R., Ten Eyck, G. A., Ginn, J.C., Ellis, A. R., Brener, I. and Sinclair, M. B.,
`“Micrometer-scale cubic unit cell 3D metamaterial layers,” Adv. Mater., 22, 5053-5057 (2010).
`
`[13] Burckel, D. B., Resnick, P. J., Finnegan, P. S., Sinclair, M. B. and Davids, P. S., “Micrometer-scale
`fabrication of complex three-dimensional lattice+basis structures in silicon,” Opt. Mater. Exp., 5, 2231-2239 (2015).
`
`[14] Boyd, G. D., Coldren, L. A. and Storz, F. G., “Directional reactive ion etching at oblique angles,” Appl. Phys.
`Lett., 36, 583-585 (1980).
`
`[15] Lee, J.-K., Lee, S.-H., Min, J.-H., Jang, I.-Y., Kim, C.-K. and Moon, S.H., ”Oblique-directional plasma
`etching of Si using a Faraday cage,” J. Electrochem. Soc. 156, D222-D225 (2009).
`
`[16] Takahashi, S., Suzuki, K., Okano, M., Imada, M., Nakamori, T., Ota, Y., Ishizaki, K. and Noda, S., “Direct
`creation of three-dimensional photonic crystals by a top-down approach,”Nat. Mater. 8, 721 (2009).
`
`
`
`
`
`Downloaded From: http://proceedings.spiedigitallibrary.org/ on 08/10/2016 Terms of Use: http://spiedigitallibrary.org/ss/termsofuse.aspx