ArticlePDF Available

Monolithic 3D-ICs with Single Crystal Silicon Layers

Authors:

Abstract

Approaches to obtain monolithic 3D logic and memory ICs are proposed in this paper. 3-4x higher memory density with similar litho cost can be obtained with monolithic 3D memories, while benefits similar to a generation of scaling can be obtained with monolithic 3D logic by doubling the number of device layers. Well-known, manufacturing-friendly materials, process steps and device structures are used.
Monolithic 3D-ICs with Single Crystal Silicon Layers
Deepak C. Sekar and Zvi Or-Bach
MonolithIC 3D Inc., San Jose, CA 95124, USA E-mail: deepak@monolithic3d.com (Invited Paper)
Abstract
Approaches to obtain monolithic 3D logic and memory ICs are
proposed in this paper. 3-4x higher memory density with similar
litho cost can be obtained with monolithic 3D memories, while
benefits similar to a generation of scaling can be obtained with
monolithic 3D logic by doubling the number of device layers.
Well-known, manufacturing-friendly materials, process steps
and device structures are used.
Introduction
3D-ICs with monolithically stacked transistor layers provide
ultra-dense through-silicon connections and short wires. In
addition, they often utilize the same litho step to pattern multiple
layers of devices, thereby reducing litho cost. A few trends are
making monolithic 3D-ICs increasingly attractive. Fig. 1 shows
litho tool cost has increased exponentially over the past 40 years.
Next-generation litho is now quite expensive and risky. Fig. 2
indicates wire RC delays are a key bottleneck in today’s chips. A
28nm GPU, for instance, requires several times more energy for
communicating data than for computation. This trend is expected
to get worse with scaling (Fig. 2). Down-scaling requires major
changes to transistors too, with disruptive innovations such as
high k/metal gate and Finfets required every few years.
Ion-Cut Technology
Single crystal silicon for stacked device layers of Monolithic
3D-ICs can be obtained with ion-cut technology, which involves
hydrogen implantation, wafer bonding and cleave (Fig. 3). Ion-
cut has been demonstrated below 400oC [3], and is well-known
due to its two decades of use for SOI wafer manufacturing.
Monolithic 3D Memories
Flash memory, which scales faster than logic and DRAM, may
hit the limits of traditional scaling first. Toshiba, Samsung,
Micron and Hynix have all invested aggressively in polysilicon-
based monolithic 3D flash memories, and have them on their
roadmaps [4][5]. Fig. 4 shows our proposal for single crystal
silicon 3D flash memories with litho steps shared among
multiple device layers. Single crystal silicon flash memory cells
have ~5x higher mobility, better off-characteristics and lower
variability compared with poly cells, all of which help with
multi-bit storage. As shown in Fig. 4, ion-cut is utilized
repeatedly to produce multiple layers of single crystal silicon
atop peripheral circuits with tungsten wiring, following which
shared litho steps are utilized to define NAND flash memory
strings and produce contacts [5]. Fig. 5 shows memory capacity
estimations based on [5]. For a 140mm2 die, the structure in Fig.
4 can provide 256Gbit chips compared with 64Gbit and 128Gbit
for conventional scaled NAND and polysilicon-based 3D
vertical NAND flash respectively [5]. This is due to the cost
advantages of sharing litho steps across multiple memory layers
as well as the multi-bit storage capacity of single crystal silicon
cells. Maximum aspect ratios for etch and deposition in Fig. 4
are 16:1 compared to higher than 50:1 for poly-based 3D vertical
NAND flash, allowing easier manufacturing. Note that ion-cut
could be applied to any 3D flash memory with horizontal
channels, and is not limited to the exemplary architecture [6]
shown in Fig. 4.
Monolithic 3D architectures with shared litho steps have been
developed by the authors for floating-body DRAM and resistive
memories too [7]. Single crystal silicon is an enabler for these
applications. The primary concern with applying ion-cut to
memories is cost. Our analysis reveals that with re-use of
substrates, ion-cut could cost as low as $60 per layer, making it
affordable. Encouragingly, some companies in the cost-sensitive
solar industry are adopting ion-cut today.
Monolithic 3D Logic
Much of today’s 3D logic stacks utilize chips processed
separately with high temperatures that are then thinned, bonded
and connected to each other. Silicon thickness of thinned die and
misalignment during bonding are concerns with this approach.
The ITRS predicts minimum TSV diameter around 1um between
2009 and 2015, indicating difficulties scaling to small TSV sizes
needed for many applications. Monolithic 3D-ICs could be a
solution. Fig. 6 indicates the main barrier to creating high-
quality transistors at Cu/low k compatible temperatures (sub-
400oC) is dopant activation. Fig. 7 describes one approach to
overcome this problem, which utilizes recessed channel
transistors. These have been used in DRAM manufacturing since
the 90nm node, and are known to be competitive with standard
planar transistors [9]. As can be seen in Fig. 7, high temperature
dopant activation steps are conducted before transferring bilayer
n+/p silicon layers atop Cu/low k using ion-cut. The transferred
layers are unpatterned, therefore no misalignment issues occur
while bonding. Following bonding, sub-400oC etch and
deposition steps are used to define the recessed channel
transistor. This is enabled by the unique structure of the device.
These transistor definition steps can use alignment marks of the
bottom Cu/low k stack since transferred silicon films are thin
(usually sub-100nm) and transparent. Sub-50nm through-silicon
connections can be produced due to the excellent alignment.
To investigate the chip-level impact of monolithic 3D-ICs, an
open-source 2D/3D chip simulator called IntSim [10] was used.
Fig. 8 gives a description of IntSim. Fig. 9 reveals that for a
22nm 600MHz logic core, doubling the number of
monolithically stacked device layers can provide similar benefits
to a generation of scaling (2x lower power, 2x lower die size).
These advantages are due to the shorter wires provided by
monolithic 3D, which allow reduced gate (driver) sizes. Note
that Rent’s Rule based stochastic wire length distributions were
used for this analysis, as CAD tools for monolithic 3D design
are still immature today. Conclusions
Several techniques to obtain monolithic 3D chips have been
proposed in this paper. Benefits equivalent to several generations
of scaling can be obtained with the technology without incurring
the cost and risk of next-generation lithography. Wiring and
litho limitations of traditional scaling could make monolithic 3D
increasingly attractive moving forward.
References
[1] W. Dally, Throughput Computing, Intl. Conf. on
Supercomputing, 2010 [2] S. Naffziger, S. on VLSI Technology,
2011 [3] M. Sadaka, et al., Building Blocks for Wafer-Level 3D
Integration, Solid State Technology [4] J. Choi, et al., S. on
VLSI Technology, 2011 [5] R. Liu, Short Course at S. on VLSI
Technology, 2010 [6] H. T. Lue, S. on VLSI Technology, 2010
[7] US Patent # 8,026,521, MonolithIC 3D Inc. [8] Y. Saito, et
al., S. on VLSI Technology, 2000 [9] J. Y. Kim et al., S. on
VLSI Technology, 2003. [10] D. C. Sekar, et al., ICCAD, 2007.
Fig. 1: Exponential increase of litho tool cost.
An immersion litho tool costs $40M today while
other tools in the fab cost $1M-$5M.
Fig. 2: Wire RC delay and energy trends
(Left) nVIDIA 28nm GPUs [1] and (Right) AMD chips [2].
Fig. 5: Comparison with conventional NAND
flash and poly 3D NAND. Fig. 6: Monolithic 3D with conventional logic transistors
would need temperatures higher than 400oC.
Fig. 8: IntSim, an open-source 2D/3D chip simulator. Fig. 9: IntSim’s results for a 22nm 600MHz logic core.
Fig. 3: Ion-Cut process for stacking single
crystal silicon (c-Si) layers at less than 400oC. Fig. 4: Process Flow for monolithic 3D NAND flash with junction-
free charge-trap flash cells made of single crystal silicon.
n Si n Si n Si
H
H
Implant H
Top layer
Bottom layer
Flip top layer
and bond to
bottom layer
400oC anneal or
mechanical force
to cleave. CMP.
Reuse original
substrate.
Use ion-cut to form multilayer
c-Si/SiO2stack above periphery Use shared litho step to define
Form NAND string with
shared litho step
Form contacts and cell source
with shared litho steps
n+ Si Oxide Gate electrode ONO
Cu/Al Cell Source Periphery with W wiring
140 sq.
mm die
2D NAND
@22nm
Poly Vertical 3D
NAND (BiCS)
32 layers @45nm
3D Single Crystal
Silicon NAND
8 layers @22nm
Density
64Gbit
(3 bits/cell)
128Gbit
(1 bit/cell)
256Gbit
(2 bits/cell)
Aspect
ratio
60:1
16:1
Sub-400oC
possible?
Method
Single Crystal Silicon
Yes
Ion-Cut
Shallow Trench Isolation
Yes
Radical Oxidation [8], HDP
High k/Metal Gate
Yes
ALD/CVD
S-D Dopant activation
No
>750oC anneal
Contacts
Yes
Nickel Silicide
Fig. 7: (a) A recessed channel transistor (b) Process flow for monolithic 3D logic. Bottom device layer with Cu/low k does not see more
than 400oC. Through-silicon connections can be close to minimum feature size due to the thin-film process.
Operation
pJ
Integer Add
1
A register file
1mm away
26
L1/L2/L3 caches
50/256/
1000
Gate
n+
n+
p
Channel
length tunable
by depth of
recess etch
(a) (b)
Oxide
Activated n+ Si
Activated p Si
Gate Electrode
Gate Dielectric
Cu
Activate dopants
at high temperature
before layer
transfer
Implant
hydrogen
for ion-cut
H
Ion-cut for transfer
of un-patterned c-Si.
No alignment issues.
Finish well-aligned thin-film c-Si recessed
channel transistors @ less than 400oC
Bottom
device
layer
IntSim v2.0
Contains models for:
Stochastic wire length
distributions of 2D/3D-ICs,
logic gates, repeaters, chip
power as well as power,
clock and thermal
interconnect networks
Inputs:
Gate count
Die area
Frequency
Rent’s
parameters
Number of
device layers
Outputs:
Chip
power
Metal
level
count
Wire
pitches
10 metal layers per
device layer
2D
@ 22nm
2 layer 3D
@ 22nm
Comment
Average wire length
6um
3.1um
Since 3D and optimal
die size less
Average gate size
6W/L
3W/L
Since less wire cap.
Optimal total silicon
area (Footprint)
50mm2
(50mm2)
24mm2
(12mm2)
Since smaller gates,
shorter wires
Power
1.6W
0.8W
Since gate, wire,
repeater area less
... Monolithic 3D-FPGA is designed using thin-filmtransistors (TFT) in the lop layer [19]. [20] have used the unique structure of recessed-channel-transistors in top layer to study its performance. Some M3D-IC based open libraries are available currently. ...
Preprint
Full-text available
Metal inter-layer via (MIV) in Monolithic three-dimensional integrated circuits (M3D-IC) is used to connect inter-layer devices and provide power and clock signals across multiple layers. The size of MIV is comparable to logic gates because of the significant reduction in substrate layers due to sequential integration. Despite MIV's small size, the impact of MIV on the performance of adjacent devices should be considered to implement IC designs in M3D-IC technology. In this work, we systematically study the changes in performance of transistors when they are placed near MIV to understand the effect of MIV on adjacent devices when MIV passes through the substrate. Simulation results suggest that the keep-out-zone (KOZ) for MIV should be considered to ensure the reliability of M3D-IC technology and this KOZ is highly dependent on the M3D-IC process. In this paper, we show that the transistor placed near MIV considering the M1 metal pitch as the separation will have up to 68,668×68,668\times increase in leakage current, when the channel doping is 1015cm310^{15} cm^{-3}, source/drain doping of 1018cm310^{18}cm^{-3} and substrate layer height of 100 nm100\ nm. We also show that, this increase in leakage current can also be reduced significantly by having KOZ around MIV, which is dependent on the process.
... Other alternatives to avoid undesirable impacts on lower layers include the use of germanium, which has lower crystallization temperature and melting point compared with silicon, the use of pulse anneal for dopant activation, and implementation of high-κ gate dielectric [13]. The use of Ultratech approach, laser spike annealing (LSA), laser annealing [18], and ion cut technology [19] have also been investigated for minimizing thermal damage to the underlying structures. ...
Article
Full-text available
Three-dimensional integrated circuits (3D IC) technology has emerged in the past few decades, driven in part by the techno-economic difficulties of dimensional scaling and the increasing interconnect delay in microelectronics. In a 3D IC, vertical integration of multiple transistor planes, either through monolithic integration or through bonding of multiple strata offers reduced interconnect delay and enhanced design flexibility. However, vertical integration in a 3D IC also results insevere thermal management challenges. Thermal management of a 3D IC is exacerbated by the multi-layer nature of the 3D IC. Further, new components such as Through Silicon Vias (TSVs) offer opportunities for novel thermal management. This paper presents a critical review of research literature related to heat transfer in 3D ICs, focusing specifically on thermal modeling, thermal-electrical co-design and thermal management of a 3D IC. Key literature from recent years is categorized and summarized. A discussion on the future outlook for research in these areas is presented. It is expected that this review paper will be helpful for researchers in academia and industry for understanding the state-of-the-art in various areas related to heat transfer in 3D ICs.
... Among the various applications of M3DI [17][18][19][20][21], the feedback field-effect transistor (FBFET) is in the spotlight as a next-generation device [22,23]. The FBFET has a S-shape energy band caused by p-n-p-n structure for n-type devices and it works on the principle of positive feedback. ...
Article
Full-text available
The optimal structure and process for the feedback field-effect transistor (FBFET) to operate as a logic device are investigated by using a technology computer-aided design mixed-mode simulator. To minimize the memory window of the FBFET, the channel length (Lch), thickness of silicon body (Tsi), and doping concentration (Nch) of the channel region below the gate are adjusted. As a result, the memory window increases as Lch and Tsi increase, and the memory window is minimum when Nch is approximately 9 × 1019 cm-3. The electrical coupling between the top and bottom tiers of a monolithic 3-dimensional inverter (M3DINV) consisting of an n-type FBFET located at the top tier and a p-type FBFET located at the bottom tier is also investigated. In the M3DINV, we investigate variation of switching voltage with respect to voltage transfer characteristics (VTC), with different thickness values of interlayer dielectrics (TILD), Tsi, Lch, and Nch. The variation of propagation delay of the M3DINV with different TILD, Tsi, Lch, and Nch is also investigated. As a result, the electrical coupling between the stacked FBFETs by TILD can be neglected. The switching voltage gaps increase as Lch and Tsi increase and decrease, respectively. Furthermore, the slopes of VTC of M3DINV increase as Tsi and Nch increase. For transient response, tpHL decrease as Lch, Tsi, and Nch increase, but tpLH increase as Lch and Tsi increase and it is almost the same for Nch.
Article
Monolithic three-dimensional (M3D) integration is gaining momentum, as it has the potential to achieve significantly higher device density compared to 3D integration based on through-silicon vias. M3D integration uses several techniques that are not used in the fabrication of conventional integrated circuits (ICs). Therefore, a detailed analysis of the M3D fabrication process is required to understand the impact of defects that are likely to occur during chip fabrication. In this article, we first analyze electrostatic coupling in M3D ICs, which arises due to the aggressive scaling of the interlayer dielectric (ILD) thickness. We then analyze defects that arise due to voids created during wafer bonding, a key step in most M3D fabrication processes. We quantify the impact of these defects on the threshold voltage of a top-layer transistor in an M3D IC. We also show that wafer-bonding defects can lead to a change in the resistance of interlayer vias (ILVs), and in some cases lead to an open in an ILV or a short between two ILVs. We then analyze the impact of these defects on path delays using HSpice simulations. We study their impact on the effectiveness of delay-test patterns for multiple instances of IWLS 2005 benchmarks in which these defects were randomly injected. Our results show that the timing characteristics of an M3D IC can be significantly altered due to coupling and wafer-bonding defects if the thickness of its ILD is less than 100nm. Therefore, for such M3D ICs, test-generation methods must be enhanced to take M3D fabrication defects into account.
Chapter
The eight chip technologies selected for the 2012 perspective on CHIPS 2020 advanced, with the exception of single-electron IC’s. Three of them merged into major cooperative R&D: The S3S initiative stands for Silicon-on-Insulator, 3D integration, and Subthreshold MOS, which we treated then and treat again extensively, particularly in differential logic.
Chapter
As the down-sizing of transistors has arrived at fundamental and practical limits, the technology direction with the largest potential for progress is the integration of transistors in the 3rd dimension on top of each other, maintaining and using the quality of monolithic, crystalline silicon in all successive transistor layers . After decades of exploratory research, monolithic 3D integration is now ready for cost-effective, large-scale implementation of nanoelectronic systems. It offers the largest gains in transistors-per-chip, it solves the on-chip interconnect and communication gridlock and thus the energy, speed and bandwidth problems. It opens a new era of effective industry networks for the sustained growth of the nanoelectronics economy. Monolithic 3D is already being adapted for mass production, in the non-volatile memory segment-3D NAND, and it can be expected that the other segments of the semiconductor industry will follow.
Conference Paper
Reported for the first time is proof-of-concept multi-stacking of memory wafers based on low-temperature oxide wafer bonding using novel design and integration of two types of ultra-fine-dimension copper TSV interconnects. The combined via-middle (intra-via) and via-last (inter-via) strategy allows for the greatest degree of interconnectivity with the tightest allowable pitches and permits a highly integrated interconnect system across the stack. In combination with the successful metallization of the ultra-fine-dimension TSVs, ...
Article
The microelectronic industry has arrived at a crossroads. There is the challenge of continued Moore's Law scaling and the ever-growing consumer demand for smaller, faster electronics with extended and new functionalities. 3D integration is a promising and fast-growing field that addresses the convergence of Moore's Law and more than Moore. 3D integration offers a path for higher performance, higher density, higher functionality, smaller form factor, and potential cost reduction. Through this emerging field, new and improved technologies and integration schemes will be necessary to meet the associated manufacturing challenges. With the possibility of addressing different process flows, this paper describes 3D building blocks, including stacking technology as well as a low temperature layer transfer technology and a metal-to-metal bonding technique.
  • Y Saito
US Patent # 8,026,521, MonolithIC 3D Inc. [8] Y. Saito, et al., S. on VLSI Technology, 2000 [9] J. Y. Kim et al., S. on VLSI Technology, 2003. [10] D. C. Sekar, et al., ICCAD, 2007.
S. on VLSI Technology
  • J Choi
  • Y Saito
US Patent # 8,026,521, MonolithIC 3D Inc. [8] Y. Saito, et al., S. on VLSI Technology, 2000 [9] J. Y. Kim et al., S. on VLSI Technology, 2003. [10] D. C. Sekar, et al., ICCAD, 2007.