Professional Documents
Culture Documents
FutureFab23 Luj Reprint PDF
FutureFab23 Luj Reprint PDF
3D Integration:
Why, What, Who, When?
>> Figure 2(d). A small piece of amor- alignment, bonding, thinning and There are also other approaches that
phous silicon film is deposited with a interwafer interconnections as shown are not shown in Figure 2, such as form-
catalyst followed by either laser heat- in Figure 2(g-k). All approaches shown ing the TSVs during the front-end-of-the-
ing or rapid thermal anneal to recrys- use TSVs to form the interstrata inter- line processing (Tezzaron), stacking chips
tallize the silicon. The transistors are connections. They differ as to when with metallization on the stack side
then formed by BEOL compatible pro- the via is formed, before/during bond- (Irvine Sensors in the U.S.), stacking chips
cessing (e.g., Stanford University in ing (via-first) or after bonding (via- on silicon carrier with TSVs (IBM) or com-
the U.S.). last). In addition, a variety of bond bining various approaches.
(e) Transistors formed on poly-silicon layer types can be chosen. Four major
films layer by layer with tungsten bonding and interstrata interconnec- Technology Comparison
interlayer vias as shown Figure 2(e). tion approaches are highlighted in and Applications
After the first layer of transistors is Figure 2(h-k): We briefly discuss the advantages and
completed, an amorphous silicon film • via-last, oxide-to-oxide bonding limitations of various 3D technologies
is deposited and converted to poly-sil- (Figure 2(h)), and their potential applications; a full
icon (e.g., MATRIX SEMI in the U.S.). • via-last, adhesive (polymer) bonding comparison among 3D technologies is
The tungsten via can tolerate the rela- (Figure 2(i)), beyond the scope of this article. In gener-
tively high temperatures (~600 C) • via-first, copper-to-copper bonding al, all 3D technologies would offer high
needed for poly-silicon conversion (Figure 2(j)), and density component integration with
and transistor formation. • via-first, metal/adhesive redistribu- • small form factor (small size and light
(f) Transistors formed on single-crystal tion layer bonding (Figure 2(k)). weight);
silicon films layer by layer as shown • reduced packaging; and
Figure 2(f). The silicon layer can be Academic and industrial organizations • reduced power (fewer I/Os to be
bonded onto the oxide surface of a are actively developing a variety of driven).
previously fabricated transistor layer wafer-level 3D technologies, such as
by transferring the crystal silicon film Lincoln Lab, MIT, Rensselaer Polytechnic For 3D packaging technologies (Figure
from a silicon-on-insulator (SOI) Institute (RPI), the University at Albany, 2(a-c)), using known good die (KGD) can
wafer. The interstrata via is filled with Freescale, IBM, Intel, SEMATECH and provide a yield advantage. Time-to-mar-
poly-silicon and/or tungsten, enabling Tezzaron in the U.S.; CEA-LETI, ket for a new product is short because
device fabrication at relatively high Fraunhofer and IMEC in Europe; and flexibility in the assembly process
temperature (e.g., Samsung in Korea). Tohoku University in Asia. Many organi- requires less design effort for a new sys-
(g) Wafer-level BEOL-compatible 3D zations are currently evaluating compet- tem. However, the cost for high-volume
hyper-integration enabled by wafer ing wafer-level 3D technologies. production can be high because of the
testing required for KGDs and the low
throughput of pick-and-place assembly.
The SiP and PoP approaches (Figure 2(a))
are used in portable devices (e.g., cell
phones). The die-stack approach with
TSVs (Figure 2(b)) has been demonstrated
for memory stacks. The die-wafer
approach (Figure 2(c)) has the potential
for building 3D SoCs.
Transistor buildup 3D technology
(Figure 2(d-f)) can achieve the highest
density of Si transistors with wafer-level
fabrication using advanced photolithog-
raphy. Wafer-level processing can also
reduce costs for high-volume production.
However, the processing constraints (par-
ticularly the thermal budget) affect the
properties of the transistors and limit the
material choices mostly to silicon and
tungsten. The Si recrystallization
approach (Figure 2(d)) could be used for
fabrication of repeaters within intercon-
nect to alleviate interconnect delay. The
poly-Si layer approach (Figure 2(e)) is
Figure 2. Schematic Representations of Major 3D Integration Approaches: (a-c) 3D Packaging used for low performance memory. The
Technology, (d-f) Wafer-Level transistor Buildup 3D Technology and (g-k) Wafer-Level BEOL- bonded crystal Si approach (Figure 2(f))
compatible 3D Technology
can be used for an NAND flash memory
26
00 | FUTURE FAB International | Issue 23
FUTURE VISIONS AND CURRENT CONCERNS SECTION 1
3D Integration: Why, What, Who, When?
stack. Companies such as Samsung that delivery (RPI). Design and simulation MEMS, and various micro/nano-scale
manufacture cell phones and handhelds tools have been demonstrated for this 3D chemical, bio, thermal, mechanical,
see great potential for a high-density technology, such as for ECAD tools electrical and optical sensors. In the
memory stack. (Lincoln Lab, MIT, R3 Logic in the U.S.), future, it will not be just a dream that a
For wafer-level BEOL-compatible 3D switching energy, and thermal and tiny device can fly, move, see, smell,
technology (Figure 2(g-k)), the electri- mechanical simulations (Intel, MIT, and hear, taste (analyze), feel and “think”; it
cal, RF, optical, thermal and mechanical RPI in the U.S.). will be able to interact with other
behavior can be considered for each devices and with human beings and
component separately. For example, the 3D Integration Perspectives their surroundings; it will also perform
starting yield can be improved by fabri- Various 3D approaches may be com- certain functions, such as for security
cating logic and memory on separate bined to offer more flexible integration (e.g., detection of dangerous materials,
wafers with optimized materials and with even higher functionality. For exam- devices or terrorists) and healthcare
process technologies. The separate ple, stacking discrete die (or discrete (e.g., medical devices and drug discov-
wafers are then stacked using a mono- devices, such as solid-state lasers or GaN ery). Our daily lifestyle and even our cul-
lithic wafer-level BEOL-compatible pro- transistors) onto a wafer stack provides a ture may be dramatically changed with a
cess for all components, potentially pathway for the integration of compound proliferation of this 3D hyper-integra-
improving the cost and overall system semiconductor devices or analog circuits tion technology.
yield at high-volume production. with digital circuits.
Typically the bottom wafer retains its Looking forward, the first killer 3D Acknowledgments
full thickness and serves as a mechani- applications would be extremely high- We would like to thank MARCO,
cal support for the stack. Each subse- density heterogeneous memory stacks DARPA, NYSTAR and SEMATECH for their
quent wafer is then thinned to micron (e.g., NAND flash, SRAM, DRAM, FRAM support. n
scale after stacking (bonding), thus the and phase change memory) and extreme-
overall stack thickness is close to that of ly high-resolution/low-cost imagers. About the Authors
a single wafer and can be processed and Advanced portable devices will also con-
packaged with current wafer technolo- tinue to drive the mix and match of vari-
James Jian-Qiang Lu
gies. Massive (millions), short (micron- ous 3D approaches. James Jian-Qiang Lu is an associate
scale) interwafer interconnects (vias) Once 3D integration technology is professor at Rensselaer Polytechnic
provide extremely high data bandwidth mature and the manufacturing infrastruc- Institute, Troy, N.Y., with 150+ publica-
and dramatically decrease interconnect ture (such as ECAD tools, fabrication tions in micro/nano-electronics theory
delay and power consumption. equipment and standards) is in place, it is and design to materials, processing,
BEOL-compatible 3D hyper-integration likely that more ICs will be designed for devices, integration and packaging; he
is perhaps the most attractive 3D technol- general purposes, because massive pro- has particular interest in 3D hyper-inte-
ogy due to its flexibility for heterogeneous duction of 3D ICs would lower the manu- gration of infotech, nanotech and
integration of different materials, pro- facturing cost. For instance, a general- biotech systems. Email: Luj@rpi.edu.
cessing technologies and functional com- purpose 3D multi-core processor (with Ken Rose
ponents with additional benefits in cost extremely high-density memory, FPGA Ken Rose is an emeritus professor at
and performance. Many research groups and software in the nonvolatile memory Rensselaer Polytechnic Institute, Troy,
are exploring its potential applications, in the same stack) could be repro- N.Y., with research interest in VLSI
such as improving interconnect delay grammed by FPGA or software for multi- design, and has led development of
(George Institute of Technology (GaTech), ple purposes, or varying purposes, over a ECAD performance estimation tools for
MIT, RPI and Stanford in the U.S., and processor’s lifetime. CPUs and memories. Email:
IMEC and Infineon in Europe), memory As shown in Figure 1, ongoing Rosek@rpi.edu.
stacks (Cornell, RPI and Tezzaron in the research and development in 3D inte- Susan Vitkavage
U.S.), memory-processor or logic-logic gration could lead to a new paradigm of Susan Vitkavage, Ph.D., manages
stacks (IBM, Intel, Freescale, Tezzaron future technologies for hyper-integration SEMATECH’s 3D Interconnect Project to
and RPI in the U.S.), signal processing cir- of infotech-nanotech-biotech systems, build infrastructure and define a
cuits (GaTech and North Carolina State enabling extremely high functionality, roadmap for 3D. Prior to joining SEMA-
University in the U.S.), field pro- high performance and small size and TECH, Dr. Vitkavage spent 15 years as a
grammable gate arrays (FPGAs) (MIT and weight with very low cost. Different com- senior manager for Bell Laboratories,
the University of Minnesota in the U.S.), ponents optimized for energy/power where her group was responsible for
imagers (Lincoln Lab and RTI can be integrated, such as processors, advanced interconnect R&D. Email:
International in the U.S.), mixed-signal memories, wireless communications, susan.vitkavage@sematech.org.
and RF applications (RPI) and 3D power special functions of nano-devices and
www.future-fab.com | 27
00