Fixed-Outline 3-D IC Floor planning with TSV Co-Placement

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 04 Issue: 06 | June-2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 224
Fixed-Outline 3-D IC Floor Planning With TSV Co-Placement
R. Priyadarsini1, C.H Kavya2
1,2Academic consultant , Department of ECE,SPMVV, Tirupati, Andhrapradesh
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - We describe, in this paper, a new digital input
output power configurable PAD for a wafer-scale-based rapid
prototyping platform for electronic systems. This wafer scale
platform includes areconfigurablewafer-scalecircuitthatcan
interconnect any digital componentsmanually depositedunits
active alignment-insensitive surface. The whole platform is
powered using a massive grid of embeddedvoltageregulators.
Power is fed from the bottom side of the wafer using through
silicon. The CPAD can be configuredtoprovideCMOS standard
voltages of 1.0, 1.5, 1.8, 2.0, 2.5, and 3.3 V using a single3.3 V
power supply. The digital I/O includes transistors sharing and
is embedded within the regulation circuitbycombiningit with
a turbo mode that insures high-speed operation. Fast load
regulation is achieved with a 5.5-ns response timetoa current
step load for a maximum current of 110 mA per CPAD. The
proposed circuit architecture benefits from a hierarchical
arborescence topology where one master stage drives 16
CPADs with a very small quiescent current of 366 nA. The
CPAD circuit and the master stage occupy a small area of
0.00847 and0.00726 mm2, respectively, in CMOS 0.18-μm
technology.
Key Words: CMOS technology, configurable voltage
reference, low-dropoutregulator,prototypingplatform,
wafer-scale system, wafer IC
1. INTRODUCTION
Today electronic systems integrate increasinglycomplex
components with demanding constraints on size, power
efficiency, and time-to-market. There have been previous
efforts to overcome the design, prototyping, and debugging
costs of high-end electronic systems,butnonehassucceeded
in all the areas needed to revolutionize system design,
prototyping, and debugging. Components, such as field
programmable interconnect chips (FPICs), are known to be
capable of reconfiguring the interconnections of an
electronic system [1].Nevertheless, a printed circuit board
(PCB) must be fabricated to match the layout pattern of the
required components for each system. Furthermore, in the
case of high-end electronic systemsthatrequirehigh-density
routing, several FPICs must be used due to their limited
capabilities [1].A novel platform made up of an active
reconfigurable wafer-scalecircuit,WaferBoard,hasrecently
been introduced for rapid prototyping of electronic systems
[2]. This platform interconnects any user integrated circuit
manually deposited on its active alignment-insensitive
surface, Wafer IC, by the system designer. This active area is
densely populatedwithover a millionsmall conductingpads,
called nano pads, and covered with an anisotropic
conductive film (z-axis film), formed with high density
micro-metallic fibers, in order to insure a good electrical
contact and to protect the Wafer IC surface.Asquarearrayof
32×32 Unit-Cells, each with a group of4×4 nano pads forms
article image (Fig. 1). The article image is photo-repeated to
create the wafer-scale circuit and with inter-reticlestitching
for connections between reticle images.
Each nano pad can be in contact with any type of user
integrated circuit ball and must, therefore, be configured as
floating, a digital input/output (I/O), a power supply, or a
ground. The configurable PAD (CPAD) is internal circuit
block that, when configured as a power supply or I/O,
provides a rang stable and regulated VDD to its connected
load. To support a large of digital user integratedcircuit ,the
CPAD can be configured to one of the nominal VDD standard
levels: 1.0, 1.5, 1.8, 2.0, 2.5, or
3.3 V, with the reference voltage internally generated rather
than externally generated as in a FPGA [3]. The CPAD also
includes a contact detection mechanism based on a user
integrated circuit ball coveringmorethanone nanopad. The
top side of the Wafer IC must be free of anyothermechanical
or electrical structures to ensure good electrical contact
between user integrated circuit balls and the z-axis film
wires. All external connections, such as power supplies and
I/O signals are carried out by through silicon vias (TSV) to
free the top side of the Wafer IC of any components or
mechanical connectors. The connection betweenthedesired
pins of two or more user integrated circuit is accomplished
via a fault-tolerant reconfigurable interconnection network,
called Wafer Net. This multidimensional mesh structure
links a number of CPADs together in each physical
direction(N-S-E-W) at near intra-chip-density[1].The Wafer
IC is a photo-repeated assembly of more than 1.2 million
CPADs. Each CPAD has to integrate all possible
configurations, meaning that an embedded regulator has to
be placed in all of them, as well as all other functionalities.
The available space for one CPAD is approximately 0.009
mm2,which makes the silicon area the most important
constraint for any integrated solution. To providestableand

configurable VDD to each userintegratedcircuit ball,voltage
regulators are distributed over the entire Wafer IC,
one per nano pad. It is known that dc-dc converters and
switching regulators can provide high power efficiency of
∼95% [4]. Nonetheless, leakage current due to constant
switching reduces the possibility of wafer-scale integration.
Moreover, such regulators often require off-chip capacitors
and inductors, which are not possible because the Wafer IC
top surface is reserved for user integrated circuit . Linear
regulators are more appealing and easier to integrate on-
chip within a small area despite their lower theoretical
maximum power efficiency, e.g., 75.8% for a 3.3 V–2.5 V dc-
dc conversion. A typical topology for a CMOS linear voltage
regulator uses a common-source output stage.Low-dropout
voltage can be achieved with thisdesign withouttheneed for
gate-overdrive[4]. The PMOS transistor, acting as the pass
device, allows it to be turned on with very large VGS and
reduces the required silicon area for a maximum desired
output current compared to an NMOS pass device. On the
downside, this topology offers poor noise rejection and its
bandwidth is limited by the low frequency amplifier
feedback [4].A solution to improve the transient response of
the linear regulator is to include a differentiator in the
feedback .Although good results and performances can be
obtained, large integrated capacitance of 2.5 pF and
resistances are needed[5]. This results in a total silicon area
of 0.29 mm2 for a similar CMOS technology, accordingtothe
authors in [5], which does not meet the area requirement.
Another approach to improve the bandwidth uses an
impedance-attenuator buffer that presents good
performances with a maximum delivered currentof200 mA,
a quiescent current of 20 μA, and a transient response of
approximately 100 ns [6]. However, the use of integrated
capacitors and resistors requires ten times the available
area, even when technology scaling is applied on the
published results of [6].Another technique uses multiple
small gain stages to enhance the transient response of the
low drop out (LDO)without introducinglowfrequencypoles
[7]. However, this solution would require more than the
available silicon area for a single CPAD while providing only
50 mA. Nested Miller compensation can also be used for
frequency compensation by using current buffer stages [8].
In counterpart, this architecture requires several integrated
passive components, such as capacitors and resistors. Other
approaches were proposed in[9] and [10] with solution 10
times the available silicon area for the CPAD. Hazucha et al.
[4] introduced a master-slave topology to minimize the
silicon area with fast load regulation
using current buffer stages [8]. In counterpart, this
architecture requires several integrated passive
components, such as capacitors and resistors. Other
approaches were proposed in[9] and [10] with solution 10
times the available silicon area for the CPAD. Hazucha et al.
[4] introduced a master-slave topology to minimize the
silicon area with fast load regulation. Exploiting this
approach within the Wafer IC would result in a
ten-fold reduction of the silicon area and power
consumption requirements, for a specific circuit, when
referring to a Unit-Cell with 16 CPADs. Fig. 1 shows a
hierarchical architecture where the master stage, one per
Unit-Cell, shares control signals with a 4 × 4 nano pad array.
The architecture reported in [4] separates the regulation
into a master-slave topology dissociatingtheslowvariations
from the fast ones (Fig. 2).A primary low bandwidth control
loop tracks a reference voltage along an Op Amp to
counteract any slow variation in the desired output. The
second loop, with a much higherbandwidth, isaccomplished
by a current feedback within the fast load regulation (FLR)
block. Despite very good performances ,the silicon area of
this design reported in [4] is anissue,0.008mm2fora 90-nm
CMOS technology (same available area for a CPAD in a 180-
nm technology), and moreover, the area of the integrated
capacitor is more than ten times higher thantheLDOone.To
our knowledge, no circuit structure exits in the literature
that meets our silicon area requirements. Regulators use a
voltage reference that must be internally generated to
establish the predefined output according to the nano pad
load requirements. External generation ofvoltage reference,
such as used for FPGA I/O banks, would not be a suitable
solution becauseofthenumerousdifferentvoltages required
(1.0, 1.2, 1.5, 1.8, 2.0, 2.5, and 3.3 V). Moreover, external
generation of needed power supplies would require more
silicon area than available and extra routing from a TSV to
every nearby nano pad as well as mechanically weakening
the wafer itself. The configurability of the internal voltage
reference could be achieved using a digital-to-analog
converter paired with a band gap reference voltage (BGR)
using a vertical PNP parasitic transistor [11], [12].However,
these circuits focus on ultralow temperature deviations of
the voltage reference, and the nano pad only supplies power
to digital user integrated circuit and can, therefore, tolerate
an overall ∼10% voltage

Variation at their power supply pin [4]. This large variation
allows us to focus on more important issues with regards to
wafer-scale integration,suchasconfigurability,fasttransient
response quiescent current, and silicon area
A distributed linear regulator can regulate a loadbytracking
a reference voltage. This reference must be as independent
as possible of any variation related to the power supply,
temperature, and fabrication process. Typical BGR uses an
operational amplifier (Op Amp)thatensuresequalityof both
differential inputs [13]. Another typical BGR uses a cascade
current mirror with PNP transistors to produce a
proportional to absolute temperature (PTAT) current.
Temperature independence is achieved by combining this
PTAT current with a complementary to absolute
temperature (CTAT) structure using a third transistor.
Others use self-bias symmetrically matched current mirror
instead of an Op Amp to reduce the circuit needed silicon
area [14].Despite the fact that these approaches are simple
and very efficient with regards to performance, they use
large resistances and parasitic diodes, making them
irrelevant for wafer-scale integration. For instance, in [13],
1.2-μm CMOS technology was used to implement the design.
The required silicon area was 1 mm2, which, with the scale-
down factor of (1.2μm/0.18 μm)2, is equivalent to 0.0225
mm2 or almost3 times bigger than the available silicon area
for an entire CPAD (0.00847 mm2). Layout optimization
would allow silicon to use efficiency but would still leave no
free space for all other required circuits. Many aspects need
to be considered when building an I/O, such as process,
voltage, and temperature variations. These variations
directly affect the driver output impedance and degrade I/O
performance [15]. The I/O can be either controlled by a
current that is more sensitive to noise or by a voltage that
offers good noise immunity. A first topology is a simple
programmable I/O driver that has the ability to calibrate the
pull-up and pull-down current drives with a combination of
PFET and NFET that can be programmed in the right
proportion .A second more complete topology uses
additional pull up and pull-down transistors and poly
resistances that flatten output impedance response [15].
However, such topology does not meet the design
requirements with regard to available siliconarea,dueto the
complexity and number of integrated resistances and
transistors. To attain configurability, a similar approach to
that available in some FPGA where an I/O voltage banks
strategy is used with their external power supplies and
voltage references would not be suitable for the Wafer IC.
Several topologies exist to generate a voltage reference,
voltage regulator, and I/O pads, but they are not amenable
for the Wafer due to their large silicon area and quiescent
current requirements. Given that theobjectiveistointegrate
one embedded CPAD per nano pad, and that a Wafer IC
includes1 245 184 nano pads, sharing maximum circuits
within a Unit-Cell would considerably help to improvethose
concerns. We propose, in this paper, a configurable PAD
suitable for wafer scale integration, which integrates a fast
load regulator that is merged with a highspeeddigital I/Oby
using transistors sharing and a turbo mode. Moreover, the
CPAD achieved the best figure-of-merit for voltage
regulation while offering a wide range of operation and has
turn-off capabilities for low power consumption. The
remainder of this paper includes, in Section II, an
introduction of the Wafer IC structure,whichdetermines the
architecture for the distributed linear regulators, voltage
reference ,I/O, contactdetection,andthestrategytoimprove
overall quiescent current. The proposed circuits and their
characteristics are also depicted in Section II. A layered
structure that can feed current from a small PCB (mm-
scale)to the active silicon areas (μm-scale) from the bottom
of the wafer is also reported in this section. In Section III, a
test chip fabricated in a 0.18-μm CMOS technology is
presented, and demonstrates the benefits of the proposed
CPAD topology.
Experimental results show that the proposed solutions for
the CPAD are reliable, efficient, and fit Wafer IC
requirements. These results are detailed and discussed, and
performances of the proposed circuits are compared with
other published works.
2. OVERVIEW OF THE WAFERIC AND DESCRIPTION
OF THE PROPOSED POWER DISTRIBUTION
SYSTEM
The Wafer IC benefits of the wafer-scale circuit
programmable capabilities to interconnect user
integrated circuit randomlydepositedonitssurface,but
that creates several design challenges to power these
multiple user integrated circuit . This section briefly
describes the Wafer IC structure and its related power
and physical requirement as well as describing
solution for power distribution and regulation specific
to the prototyping platform.It shows an adequate
solution that suits wafer-scale integration.
2.1 Proposed Wafer IC Topology and Design
Requirements
Fig. 1 shows a hierarchical view of the Wafer IC, which has
been implemented in a mature and low-cost seven metal
layer CMOS 0.18-μm technology. Its active circuit is
composed of76 photo-repeated reticle image tiles within a
200-mm wafer (Fig. 1). Each reticle image includes a 32×32
array of 560×560 μm2 Unit-Cells. Each Unit-Cell cansupport
two 250-μm diameter user integrated circuit balls with an
800-μm pitch. It is also important to have 4 × 4 nano pads
per cell to ensure good electrical and mechanical contact
with the user integrated circuit balls, deposited onthez-axis
anisotropic conductive film and conductive nano pad [16].
The Unit-Cell circuitries include4 ×4CPAD,thecontrol logic,
the Wafer Net, and the analog circuits [1]. The CPAD is the
largest element at 77 μm × 110 μm,consuminginaggregated
almost half silicon area. Each CPAD can provide 110 mA to a
user integrated circuit ball load. The specified sizes have

been done accordingly to fit the requirements of the 4 × 4
array of nano pads explained earlier, and it is also a tradeoff
of silicon area with the other necessary digital circuitries
essential to a Unit-Cell. A contact detection mechanism is
also implemented to locate physical connections to a user
integrated circuit pin by detecting shorts created between
two or more adjacent nano pads by a user integrated circuit
ball (Fig. 1). This is done by addressing pairs of neighboring
CPADs with appropriate pull-up and pull-down signals and
activating the proper detection circuits. The power supply
arborescence structure combined with the size of thepower
rails makes all power supplies uneven across the wafer, due
to parasitic resistances, processvariations,the largenumber
of supplied user integrated circuit , and the random
placement of these user integrated circuit over the surface.
These physical and structural constraints make the ground
and power grids very noisy. The external voltage regulators
feeding voltage through several TSV could suffer severe
voltage drops and noise at the user integrated circuit ball
due to parasitic resistances of the TSV and the on -silicon
power and ground grids. A linear voltage regulator closer to
the load, within each nano pad with virtually no distance to
the regulated load, is more amenable.Embedded distributed
voltage regulators are proposed to alleviate against dc
variations and all frequency noise within the power grids
and substrate.
2.2. Power-Supply Arborescence Topology
A ground and two other power supply rails, 1.8 and 3.3 V
(typical power supply voltages for a 0.18-μm CMOS
technology),are distributed uniformly over the 200-mm
wafer. The Wafer Net logic core is supplied with 1.8 V,witha
maximum current of 5 A split among a maximum of four
reticule images. The CPAD and all other analog circuits are
supplied with the3.3 V rail for the same number of reticle
images but with5 times the power of the 1.8 V rail.
The Wafer IC power-supply circuit has an arborescence
structure alongside a single power source as the root, and
the embedded regulators in each CPAD as the leaves (Fig.
3).Powering up the complete wafer is accomplished by an
ac/dc converter fed by a PCB that interconnects 21 Power
Blocks with a ground and a 12 V supply. Each Power Block
performs a dc-dc conversion down to 1.8 and 3.3 V and
provides powers to the Wafer IC through its bottom surface
to keep its active surface exempt of any components. Each
Power Block powers up to four reticule images throughouta
massive grid of TSVs. An on-silicon metal griddistributesthe
necessary voltages through all Unit-Cells and nano pads.
Decoupling capacitors cannot be placed on the topside of
The Wafer IC because the link between user integrated
circuit balls is digital. More over, placing a decoupling
capacitor on the same nano pads the user integrated circuit
ball would not be possible. Integration of sufficient
capacitance is also impossible due to siliconarea constraints
imposed by the Wafer IC. Consequently, the chosen
architecture needs to deliver a regulatedvoltagewithoutthe
benefit of adding capacitors. Moreover, it is necessary to
distribute a stable voltage level to the powered CMOS user
integrated circuit deposited on the prototyping platform, in
relation to VDD fluctuations, process, and temperature
variations. Typical linear regulators use an Op Amp to
ensure that VOUT tracks V ref across all these variations.
However, process, temperature,and VDDvariationsareslow
(<kHz) while load current spiking is very fast (MHz),Several
linear regulator topologies exist and offer a wide range of dc
load regulations, output impedances, quiescent currents,
silicon areas, decoupling capacitances, and transient
responses. The requirements listed in Table I are the fore
most objectives regarding required output range, maximum
current, and silicon area. We adapted the which makes this
approach unappealing.
The master-slave approach proposed in [4] in order to
minimize silicon area by sharing and combining a maximum
of circuits. In Fig. 2, the master stage (Unit-Cell)
encompasses the low bandwidth loop and there is one FLR
block per nano pad for the higher bandwidth loop.
The silicon area has been reduced by a factor of four, and a
novel digital configurable output is integrated to the

architecture proposed in [4] by reusing the pass-transistor
and the fast-loop feedback.
This section estimates
2.3 Through-Silicon Via Density and dc IR Drop
Analysis
The dc performances of the proposed
power supply arborescence anddeterminesifitcouldsupply
sufficient power to a user integrated circuit . Every stage
introduces a dc voltage drop that can be modeledby discrete
serial resistances. An overall dc drop of 150 mV from the
root to the CPAD power supply is acceptable,leavingenough
headroom for all circuits to work properly. The main
backside PCB is considered as perfect power and ground
plane due to the thickness of the metal lines. The TSV
resistance was measured in [16]; with a value of ∼11 m_ per
TSV. To determine the adequate TSV density and the
equivalent resistance of the on-silicon
Metal grid, modeling and simulation were performed using
COMSOL Multi physics [17]. For the purpose of these
Simulations , four power pins were each in contactwithfour
nano pads. Each pin reduces themaximumavailablecurrent,
by a total of 400 mA per pin, which represents a powerful
microprocessor activity. The results are summarized in
Table II. The thickness of the CMOS technologymetal layeris
less than 1 μm, therefore it cannot carry a lot of current. To
enhance the capabilities of the power grids, the density of
TSVs/mm2 must be increased, but the TSV density islimited
by the mechanical strength of a 200-mm wafer, the more
holes it contains, the more fragile it is. The metal density is
defined as the ratio of metal width to metal pitch. Fig. 4 is a
representation of ourmodel. Thismetal densityhasanupper
bound because the routabilityof the Wafer IC logic must be
maintained using six metal layers [1]. A TSV density of0.25
mm2 with a metal wire density of 6/30 (width/pitch) was
selected for the actual implementation. The maximum dc
overall voltage drop calculated with this setup, using
COMSOL Metaphysics, is 90 mV, which is close to half the
maximum acceptable dc voltage drop described earlier in
this section.
2.4. Master Stage Topology
Fig. 5 depicts the transistor level view of the proposed
Master stage topology. All transistors are high voltage
devices, except transistor M50, which is a native (low
threshold)component to help improving the outputrange.It
consists of a configurable voltage reference array (CVRA)
that feed san Op Amp with a command (VREF) implemented
in order to have a good power-supply rejection ratio in
addition to good temperature independence. The OpAmp
tracks VREF to ensurestability of VOUT0. Thedesiredoutput
voltage is obtained by supplying the FLR with the voltage
VSET generated by an N Stage Buffer. The objective of this
buffer is to isolate the operational trans conductance
amplifier (OTA) from any noise induced by the load of the
connected slave stages. Considering that the prototyping
platform needs to powers IC with standard voltage level, as
well as being compatible with the samedigital levels,thereis
no need for a configurable voltage reference with
intermediate levels. Voltages below the targeted standard
CMOS are not desired as an increase of temperature will
slightly increase the regulated voltage, helping to
compensate for the dc IR drops of the power-supply chain.
The proposed architecture uses a beta multiplier
architecture to provide a current, IREF, that ideally only
depends on transistor parameters. This current is then
duplicated into a transistor-based array of voltage divider
sto provide the required reference standard voltages of
1.0,1.5, 1.8, 2.0, and 2.5 V [18], [19]. Given that the master
stage will be instantiated 77 824 times, the very low
quiescent current and the small silicon area of 0.0014 mm2
in a CMOS0.18-μm technology, offer an attractive solution to
achieve configurability since no vertical PNP parasitic
transistors or other large biasing current is needed.
This approach uses two different voltagedividersinorderto
achieve the targeted references. The first one can be
expressed by (1) assuming M14 is saturated,M15isintriode
region, and their drain currents are equal to IREF generated
by M13
The second voltage divider technique cascades two
saturated
transistors. Assuming drain current ID16 equals ID17, the
reference voltage can be expressed as follows:
where IREF is given by (3) and KPn is the trans conductance
parameter and M the transistor size ratio of M11 and
M10.Equation (4) shows the temperature dependence (dT)
of IREF. As temperature increases the mobility decreases
leading to a reduction of the trans conductance parameter
butan increase of R1. Those two effects tend to cancel each

other improving the temperature independency but still
slightly
CTAT according to post-layout simulations [19]
The small signal analysis of the current mirror in the first
stage of the proposed master stagearchitecture(Fig.5)leads
to the expression of the output impedance seen looking into
the drain of M6, which is mirrored in M13 as
Equation (7) gives the trans conductance gm10 from the
basic of a beta-multiplier where Vgs10 = Vgs11+ ID11R1,
given by (6), and shows the temperature dependence of the
output current from the mirror current stage. From Fig. 5
Table -1: Sample Table
Replacing gm10 by its expression in (5), the output
impedance becomes
The output impedance ROUT of the current mirror depends
on the value of the resistance R1.An Op Amp structure was
selected with power consumption and silicon area
requirements in mind. The open loop gain and bandwidth
are not critical since temperature and process variationsare
slow, a gain around 40 dB is sufficient [4].
The output, VAMP, is connected to NB and generates an
Output command, VSET that swings from 0.5 to 2.5 V. VSET
is connected to each FLR slave stage in each nano pad and
Duplicated FLR is integrated into the master stage.
Considering that all FLR stages are identical, the output
voltage at the nano pad is equal to VOUT0.
To ensure low-power consumption of all master stages
Implemented on the wafer, kill switches were added to all
Described modules, controlled by an ON_MASTER signal.
The CVRA is put offline by shorting all gates of PMOS
Transistors connected to power-supply, to 3.3 V. The same
Approach is used on NB and FLR modules by disconnecting
Appropriate transistors. The disconnection of the OpAmp is
done by disrupting the biasingcurrentofthedifferential pair
with M12, the gates of M4, M5, and M13 are then shorted
to3.3 V ensuring the complete halt of the Op Amp. Failure to
do this would result in the summation of 77 824 times a
biasing current of several hundred microamperes. E. Slave
Stage — the CPAD The proposed slave stage is shown in Fig.

6. It combine s several distinctive circuits to fulfill the
required features in the CPAD, all driving a single nano pad
opening when acting as an output: configurable regulator,
configurable digital output, variable digital input, contact
detection, and ground. All transistors are high voltage
devices, only M55 is a native device to help improving the
output range.
The configurable FLR is brought from transistors M51 to
M55, where the current is pushed by M53. M52 and M53 act
as a super source follower while M54 as a common gate
amplifier. The command VSET defines the desired non
inverting output voltage with an additional threshold. When
the nano pad load sinks no current, M53 barely turns on to
accommodate the biasing current provided by M55. As the
current demand increases, VOUT drops and M54 starts to
turn off redirecting the fixed biasing current through M52,
which pulls down the gate of M53 to readjust the current
demand. When not in operation, the entire regulationcan be
turned off, reducing the quiescent current to virtually
nothing. This is accomplished through the signal OE that
shorts the gates of M51 and M54 to the appropriate voltage,
placing the output in high impedance and interrupting all
other current flows. The standard CMOS voltage level
requirements for regulation and digital output are thesame,
linking both functions.
Moreover, the restricted silicon area suggests thepossibility
of transistor sharing. Using the configurable voltage
regulator to define the input or output voltage levels when
configured as a digital I/O would result in a significant
surface gain. However, the limited capabilities of the
regulator, with regard to rising and falling times, are limited
due to the size of transistorM53. The suggested architecture
to accelerate transition time is shown in Fig. 6, where the
digital I/O is embedded into the regulator. A differential pair
is added to boost the response time of the power transistor
gate. Transistor M68 was designed to pull down VTURBO to
the ground, opening M53 to the fullest to rapidly boost the
output. Once VOUT reaches VSET,whichisalwayssmallerby
a threshold voltage VTH, the differential pair ceases to
operate and regulation takes over for the last few milli volts.
A custom NAND gate, made with transistorsM61–M66,
ensures that this circuit operates with appropriate timing
and is disconnected with the ON_IO signal when the
I/O function is not needed. Digital input is performed by a
single level shifter(M74–M80) that takes any voltages from
1.0 to 3.3 V, converts it to a clean 1.8 V and propagates itinto
the Wafer IC digital interconnection network. Due to lack of
silicon area, the digital input was calibrated to offer a
symmetric duty-cycle at 1.8 V.A more complex solution
larger silicon area consumptionto ensuresymmetrical signal
at any voltage would not be suited for the Wafer IC
requirements.
2.5. Layout of the Proposed Test chip
The fabrication of a complete 200-mm Wafer is costly, not
only in money but also in development time.Basicprinciples
and feasibility of the Wafer Net, configurability via a JTA
Glink, and all CPAD associated features were tested with a
low -cost and quick turnaround solution.Ashuttle runwitha
test chip several times smaller than the actual full-scale
prototyping platform was realized. Fig. 7 is a micro
photograph of the proposed layout ofthetest chipconsisting
of1024 nano pads and 64 master stages in an 8×8 square
Unit-Cell matrix.
3. MEASUREMENT RESULTS
3.1. Embedded Regulators dc Characteristics
The FLR was designed to achieve a maximum current of
over 110 mA per CPAD, validated with destructive
measurements .Fig. 8 shows the V/I curve over different dc
loads. The layout of the test chip limits the dc
characterization to a maximum current of ∼20 mA, because
it could not benefit from the power supplied by TSVs as
available in a real Wafer IC. As a result, the last metallization
layer is
shared with the nano pads, the output pads, and the power
grids of the chip itself. It is important to note that the
produced voltages, under a no-load condition, always seem
to be tens of millivolts higher than the expected value. The
fact that the nominal voltage was chosen to be at a50-mA
current load, half of the maximum current load of the CPAD,
explains this deviation. The OE signal stops any current
source when activated, leaving a very low quiescent current
of5.85 μA for an entire Unit-Cell or the equivalent of 366 nA
per CPAD.
3.2. Embedded Regulator Transient Response to a
Load Step
The accurate validation of the regulator requires a transient
load with a rising edge of few nanoseconds, which was
determined through calibration measurements.Fig.9shows
the transient response using an external 1-MHz load clock
with a 50% duty cycle. Rising and falling times were <8 ns

with20-mA amplitude due to the test chip limitations.
According to Fig. 9, the dynamicoutputimpedanceiscloseto
1 _,
with drops of 1% of the configured voltage. Notice that a
user integrated circuit ball should be in contact withseveral
nano pads concurrently,ideally four or more reducing the
output impedanceaccordingly.
3.3. Power-Supply Rejection
The fabricated test chip was tested with a 400-mVpp
sinusoidal power supply added to a dc offset of 3.3 V at 1
MHz and1 kHz, without any decoupling capacitance at VDD.
Figs. 10 and 11 show that, even under stressful conditions,
the output voltages remain stable with about 20 dB of
power-supply rejection for those two frequencies.
3.4 Embedded Digital I/O Characteristics
Fig. 12 shows VOUT when configured as a digital output.
These results were obtained by cascading two CPADs, one
Configured as an input and the second one as an output.
The results presented were probed in a test bench with a
high parasitic capacitance (∼nanoFarads),whichresulted in
the impossibility to gather high-speed results such as those
suggested in post-layoutsimulation(∼300MHz).Despite the
high-speed incapability of the test configuration, we were
able to demonstrate the configurability of the digital
interface. The remaining function of the CPAD, such as the
ground, is
Carried out by a large NMOS transistor (M56). This
transistor is turned on by controlling circuitry of M57–M60,
when all other major functions associated with power
transistorM53are not in operation. The CPAD also includes
contact detection mechanism that allows the user interface
software, configuring the Wafer IC, to pinpoint all nano pads
connected together when a user integrated circuit ball
shorts a certain number of nano pads.This isachievedwitha
system of weak pull-up and stronger pull down
(M81–M84) alongside a custom XOR gate (M85–M92)
that would generate an appropriate signal when the inputs
of this gate are VOUT and the control signal NA_WT_TPT_0.
If two nano pads are shorted by a user integrated circuit
ball, applying the pull up to the first and the pull-downto the
second would resulting an overall ground, which is not the
expected value for the first nano pad. The detection of this
contact would result in a digital “0” signaling a short.
3.5 Performance Comparisons
The proposed configurable CPAD was implemented in a
180-nm CMOS technology. For a simulated ∼110-mA post
layout maximum current rating, the total area for one CPAD
is 0.00847 mm2, where half the area is occupied by the
output stages. When scaling previous works done in [4] to
a180-nm technology, the used silicon area becomes 0.036
mm2.Our circuit represents an overall improvement of the
silicon area by a factor of more than four times when
compared to this original version. Moreover, the proposed
CPAD integrates more functionality,suchastheconfigurable
digital I/O. The master stage requires almost the same area,
0.00726 mm2. This design does not require any on-chip
capacitance; it is meant to
Work only with parasitic capacitance conveyed by a user
integrated circuit ball
And its power grid (∼1 nH, 5 nF). The response time, TR, is
Calculated from the required capacitance CLOAD for a
specific
For the purpose of comparison with other regulators
designed with specific application or design goals, such as
Quiescent current IQ, decoupling, and maximum rating,
IMAX, we used a figure-of-merit (FOM) detailed in (10). The
smaller
is the FOM, the better is the regulator. According to Table III,
this paper achieves smaller FOM than all others designs
4. CONCLUSION
A promising platform to rapidly prototype electronic
systems was developed in our lab. A smart configurable
wafer scale active circuit allows system designers to
manually deposit integrated circuits on its surface and to
power and interconnect them all together. We focused, in
this paper, on the power-supply voltage distribution of this
structure, which is fed through a dense grid of through
silicon combined with an arborescence power-supply
topology allowing minimal dc voltage drops and the
embedded digital output that shares theLDOtransistors. We
demonstrated a CPAD in 180-nmCMOS technology with
configurable standard voltage for voltage regulation, aswell

as for digital input and output. We achieved a fast response
time of 5.5 ns and the best figure of-merit, which takes into
account decoupling capacitance, maximum current,
quiescent current, response time, and output deviation. The
proposed pads fulfill other applications, such as ground and
contact detection between similar pads. The silicon area of
the whole CPAD and the master stage with no integrated
capacitor are 0.00847 and 0.00726 mm2, respectively.
The 366-nA quiescent current makes this CPAD suitable for
the Wafer IC.
REFERENCES
[1] E. Lepers, O. Valorge, Y. Basile -Bellavance, N. Laflamme-
Mayer,Y. Blaquière, and Y. Savaria, “An interconnection
network for a novelreconfigurable circuit board,” in Proc.
2nd Microsyst. Nano Res. Conf.,
Oct. 2009, pp. 53–56.
[2] R. Norman, O. Valorge, Y. Blaquière, E. Lepercq, Y. Basile-
Bellavance,Y. El-Alaoui, R. Prytula, and Y. Savaria, “An active
reconfigurablewafer-scale circuit board,” in Proc. Joint IEEE
NEWCAS TAISA Conf.,Jun. 2008, pp. 351–354.
[3] Virtex-6 FPGA Data Sheet: DC and Switching
Characteristics, XilinxCorporation, San Jose, CA, Sep. 2010.
[4] P. Hazucha, T. Karnik, B. A. Bloechel, C. Parsons, D. Finan,
and S.
Borkar, “Area-efficient linear regulator with ultrafast load
regulation,”IEEE J. Solid-State Circuits, vol. 40, no.4,pp.933–
940, Apr. 2005.
[5] R. J. Milliken, J. Silvia-Marinez, and E. Sanchez-Sinencion,
“Full onchipCMOS low-dropout voltage regulator,” IEEE
Trans. Circuits Syst.I, Reg. Papers, vol. 54, no. 9, pp. 1879–
1890, Sep. 2007.
[6] M. Al-Shyoukh, H. Lee, and R. Perez, “A transient-
enhanced lowquiescentcurrent low-dropout regulator with
buffer impedance attenuation,”
[7] M. Ho, K. N. Leung, and K.-L. Mak, “A low-power fast-
transient 90-nmlow-dropout regulator with multiple small-
gain stages,” IEEE J. Solid-State Circuits, vol. 45, no. 11, pp.
2466–2475, Nov. 2010.
[8] A. Gari mella, M. W. Rashid, and P. M. Furth, “Reverse
nested miller compensation using current buffersina three-
stage LDO,” IEEE Trans.Circuits Syst. II, Exp. Briefs,vol.75,no.
4, pp. 250–254, Apr. 2010.
[9] K. N. Leung and Y. S. Ng, “A CMOS low-dropoutregulator
with amomentarily current-boosting voltage buffer,” IEEE
Trans. Circuits Syst.I, Reg. Papers, vol. 57, no. 9, pp. 2312–
2319, Sep. 2010.
[10] O. Trescases, A. Prodié, and W. T. Ng, “Digitally
controlled currentmode DC–DC converter IC,” IEEE Trans.
Circuits Syst. I, Reg. Papers,vol. 58, no. 1, pp. 219–231, Jan.
2011.

Fixed-Outline 3-D IC Floor planning with TSV Co-Placement

More Related Content

What's hot (17)

Similar to Fixed-Outline 3-D IC Floor planning with TSV Co-Placement (20)

More from IRJET Journal (20)

Recently uploaded (20)

Fixed-Outline 3-D IC Floor planning with TSV Co-Placement