DECAP

As integrated circuit (IC) technology scales, more and more

transistors are being placed within a single chip, while the clock frequency
continues to increase into the gigahertz range. The result is that large
transient currents are drawn from the power supply rails in just a few
hundred picoseconds [1] in modern custom and application-specific
integrated circuit (ASIC) chips. Meanwhile, the supply voltage is scaled with
technology to reduce overall power consumption, and as a consequence, the
circuitry becomes more prone to power-supply noise. The management and
regulation of the quality of the on-chip power supply is a major challenge [1].
The power grid, which provides Vand V (or ground) signals throughout
the chip, DDSS experiences fluctuations in value due to a variety of noise
sources. If the supply voltage noise or variation is excessively large, it may
lead to problems such as delay variation, timing unpredictability, or even
improper functionality [2]. A commonly used metric, noise budget, is defined
as the maximum allowable noise amplitude [3]. Typically, it is required to
keep the power supply noise within a certain percentage (e.g., 10%) of the
nominal supply voltage.
A common technique for reducing power supply noise and keeping the
noise within the noise budget is through the use of on-chip decoupling
capacitors (decaps). Decaps are essentially capacitors that hold a reservoir of
charge and are placed close to the power pads and near any large drivers.
When large drivers switch, the decaps provide instantaneous current to the
drivers to reduce IRdrop and Ldi/dt effects, and hence keep the supply
voltage relatively constant. As shown in Figure 1.2, the on-chip decap
delivers current to charge up the load capacitance of the second inverter
when it switches. The supply voltage level is relatively constant at the
inverter tap point since the decap is nearby, so V is minimal.

At the 90nm technology node, the oxide thickness of a transistor

is reduced to roughly 2.0nm or less. The thin oxide causes two new
problems: possible electrostatic discharge (ESD) induced oxide breakdown
and gate tunneling leakage [9], [10]. Potential ESD oxide breakdown
increases the likelihood that an IC will be permanently damaged during an
ESD event, and hence raises a reliability concern. Higher gate tunneling
leakage increases the total static power consumption of
the chip. As
technology scales further down, with a thinner oxide, the result is an even
higher ESD risk and more gate leakage. The standard decap design using
NMOS transistors experiences these two problems and therefore becomes
rather inappropriate for 90nm and below.
While satisfying ESD reliability and gate leakage limitations, decap
designs must also meet the transient performance requirements. Since a
90nm process (or below) usually provides the capability of running at
gigahertz frequencies, the decap must respond in the order of a
hundredpicoseconds.
A new cross-coupled standard-cell design approach [11]
addresses the issue of ESD performance. The design provides a certain
amount of ESD input protection to the decap, but does not offer any savings
in gate leakage. Even worse, the new design experiences a much degraded
transient response, making it somewhat unsuitable for high-speed chips.
Modifications to the cross-coupled decap design that properly trades off ESD,
transient response and gate leakage are needed.

Another new global decap design approach, called gated decap, has
been reported [12] to control gate leakage. Based on the approach of multithreshold CMOS circuit, the gated decap is capable of saving a significant
amount of leakage current while in power-saving mode. However, the design
suffers from an oscillation problem. The lack of robustneess of the design
makes it somewhat less attractive for industrial use.
Standard-Cell Decap Layout and Placement
In white spaces, decaps are usually made of NMOS devices, as
described in the early sections. However, within standard cells, it is more
convenient to make decaps using both types of NMOS and PMOS to form a
decap filler cell, as shown in Figure 2.5. This is because the n-well is already
implemented and usually reserved for PMOS devices. Only the lower half-cell
area is for NMOS devices.

One sample standard-cell decap layout is illustrated in Figure 2.6. In

the figure, the NMOS decap occupies roughly the bottom half of the cell area,
whereas the PMOS decap is located in the n-well. The capacitor areas are the
polysilicon gates placed on top of the channel regions of the MOS transistors.
For standard cells, the height of the cell is always fixed, and the designers
can only adjust the cell width. Once the cell width is determined, the size of
the decap and the capacitance of the decap are established. Figure 2.6
implies a large decap cell (measured in cell width) with long channel
transistors.

Cross-Coupled Decap
Knowing that the standard N+P decap design for standard cells
may no longer be suitable for 90nm technology due to increased ESD risk, a
new cross-coupled decap design has been proposed [11] to address this
issue. In the new cross coupled design (Figure 3.1), the drain of the PMOS
connects to the gate of the NMOS, whereas the drain of the NMOS is tied to
the gate of the PMOS.

From the layout perspective, this cross-coupled circuit can be

seen simply as a terminal-swapped version of the standard decap. In other
words, the decap transistor areas need not to be modified, while only the
metal wire connections are modified. Thus, this new design does not require
additional area in layout, compared to the standard design.
Both transistors in this design are still in the linear region. In the
standard decap design, the gates of the transistors are directly connected to
either Vdd or Vss, depending on the transistor type. In this case, the gate of
the NMOS device is connected to V
through the channel resistance of the VDD PMOS device. Similarly, the gate
of the PMOS device is tied the channel resistance of the NMOS device and
then connected to Vss.The added channel resistance to the gate provides the
input resistance R
for ESD protection, as previously mentioned in Section 2.4. The input
inresistance can help to limit the maximum current flow to the decap so that
the voltage seen from the gate of the decap is also limited. Intuitively, the
input resistance along with the decap can be thought as a low pass filter, as
illustrated in Figure 3.2. When there is a sudden voltage jump at the power
line (Vin) during an ESD event, the voltage at the gate of the decap (Vgate)
does not increase instantaneously with the increase of Vin. Instead, the

increase of V is delayed due to the low-pass RC effect. This time gate delay
in the voltage change at the transistor gate helps to protect the gate until the
primary and secondary ESD devices are fully operational and shunt the ESD
current away. Hence, it is desirable to have Rin as large as possible from the
standpoint of ESD protection.

By simply swapping the terminal connections, the cross-coupled design adds

a considerably large Rin to the gate of the decaps, without increasing the
layout area. The tradeoff of this design is the reduced transient performance
as a decap.
Since both transistors are in the linear region, the two transistors are
on and do not limit the gate leakage current flow compared to the standard
design. Hence, no savings in gate leakage are achieved in this cross-coupled
approach. This design nicely illustrates the concept of tradeoffs between
transient response and ESD reliability when designing a decap. This crosscoupled circuit is also considered to be fairly passive and robust. In the next
chapter, this circuit will be discussed in greater detail, and modifications will
be provided to address different design tradeoffs including gate leakage. Two
of the modification circuits can achieve higher gate leakage savings and
provide comparable or better transient response, while another is better in
terms of ESD reliability.
Gated Decap
Chen et al. [12] has recently implemented a new decap structure
that saves gate leakage. The structure is called gated decap, as shown in
Figure 3.3. A control transistor is inserted between the standard NMOS-only
decap and the Vss line. The source and drain of the decap are connected to
the source of the control transistor, making the node a virtual ground
(V_GND). The drain of the control transistor is tied to the real Vss.As shown
in Figure 3.3, the substrate of the decap is still attached to Vss. There are
two major components in gate leakage: leakage current from gate to channel
(Igc), and from gate to substrate (Igb). The current Igc can be partitioned
into two: leakage current from gate to source (Igcs) and from gate to drain
(Igcd) [19]. The amount of gate leakage from gate to substrate Igb is
roughly 10x smaller than the leakage from gate to channel Igc [15] [16].
Thus, the substrate of the decap does not need to be tied through the control
transistor, and the leakage current Igb is neglected.