Professional Documents
Culture Documents
A Fully Differential CMOS Capacitance Sensor Design, Testing and Array Architecture
A Fully Differential CMOS Capacitance Sensor Design, Testing and Array Architecture
Abstract— The paper presents a fully differential capacitance as a result of device and process mismatch effects [6]. This
sensor employing the CBCM technique to map differential can be resolved by employing larger device sizes in the mea-
input capacitances to rail-to-rail differential output voltages. The surement circuit which increases the on-chip sensor footprint.
circuit has been designed for measuring capacitances in the
±20 fF range, appropriate for sensing live cells using on-chip The proposed array architecture allows the entire sensor array
microelectrodes. The paper also proposes an array architecture to be read out using a common capacitance measurement unit.
based on a shielded current routing bus that allows for a single This increases the total on-chip area available for sensing. This
measurement circuit to be shared by all the sensor pixels without configuration is also much simpler to calibrate for inter-pixel
compromising performance. This eliminates the need for individ- mismatch when compared to arrays with in-pixel measure-
ual pixel calibration. Each sensor pixel comprises 6 minimum size
digital transistors, enabling high density integration. The sensor ment. The current bus shielding preserves sensor evaluation
employs a 3-phase clocking scheme that enables gain tuning speed and improves immunity to substrate noise.
and also limits output voltage offsets. The paper presents data The integrated capacitance measurement circuit presented
obtained from 5 chips fabricated in a commercially available 2- here can be widely employed for several on-chip cell sensing
poly, 3-metal, 0.5 μm CMOS technology, each of them comprising applications including cytometry, adhesion characterization,
individual circuits measuring the substrate-coupling capacitances
of metal3 electrodes of varying sizes. The test data indicates viability monitoring and proliferation tracking [8]–[10].
successful sensor operation with a maximum sensitivity of 126 II. S ENSOR C IRCUIT D ESIGN AND O PERATION
mV/fF, a maximum achievable resolution of 14 aF and an output
dynamic range of 69.4 dB. Fig. 1(a) shows a schematic of the capacitance measurement
I. I NTRODUCTION circuit. It comprises a standard CBCM front-end (M1-M4),
The charge based capacitance measurement (CBCM) tech- a pair of complementary current mirrors (M5-M8 and M10-
nique has enabled on-chip measurement of fF capacitances M13), current subtractors (M8-M9 and M13-M14), integration
with aF resolution. The CBCM approach was primarily de- capacitors (Cint− and Cint+ ) and rail-to-rail readout buffer
veloped for interconnect capacitance characterization over a amplifiers (B1 and B2). Ci− and Ci+ represent the input
decade ago [1]. Since then it has been employed in several nodal capacitances, which vary depending upon the method
other applications including measurement of MOS device used for capacitance compensation [9] and are formed by the
C-V characteristics [2], particle detection for industrial and stray circuit capacitances, standing electrode capacitances and
biomedical purposes [3]–[6], and DNA sensing [7]. the sensed capacitance at the sensor nodes.
We present a sensor design tailored for on-chip cell moni- The CBCM unit comprises two identical pairs of minimum
toring which requires the circuit to measure capacitances over size NMOS and PMOS transistors (M1,M3 and M2,M4),
a few 10’s of fFs [8]–[10]. It extends previously reported that are switched using two non-overlapping clock signals
CBCM circuits with single-ended output configurations [3]– reset and eval. The sensing operation proceeds in three
[6] to a differential output architecture. The fully differential phases: reset, evaluation and sample. Fig. 1(b) shows the
measurement circuit compensates for parasitic capacitances timing control signals corresponding to the three phases and
associated with on-chip cell sensing, increases sensor dynamic the sensor response waveforms. During reset, Ci− and Ci+
range, and suppresses correlated noise for improved assess- are discharged to V ss through M1 and M2, and Cint− and
ment of cell phenomena [9]. The sensor circuit presented here Cint+ are reset to the common-mode voltage V cm. During
employs a 3-phase clocking scheme that allows for gain tuning evaluation, Ci− and Ci+ are charged to V dd − |Vthp | through
in accordance with the conditions in a particular cell sensing M3 and M4, where Vthp refers to the PMOS threshold voltage.
application. These conditions include sensing electrode areas The average values of the charging currents Ic− and Ic+ can
and configurations, and the dielectric parameters of the cell- be expressed as:
Ic± = Ci± · V step · f (1)
substrate interface and the growth medium.
We also propose a shielded current routing bus architecture where V step = V dd − V ss − |Vthp | and f = 1/T is the
which enables the measurement circuit to be incorporated into sensing cycle frequency. Ic− and Ic+ are amplified by the
sensor arrays. CBCM sensors are heavily prone to variability current mirrors M5-M8 and M10-M13 with gain Ai and then
subtracted by the transistors M8-M9 and M13-M14 to yield The differential sensor can be configured to compensate for
complementary difference currents ΔIo+ and ΔIo− , whose the stray and standing capacitances using different methods
average values can be expressed as: depending upon the application [6], [9]. For example, the
sensor illustrated in Fig. 1(a) compensates for both stray and
ΔIo± = ±Ai · ΔCi · V step · f + Ios± (2)
standing capacitances. The success of these schemes relies on
where ΔCi = Ci+ −Ci− is the differential input capacitance, matching between the various components in the sensor.
and Ios− and Ios+ are the offset currents in the current
III. S ENSOR A RRAY BASED O N A S HIELDED C URRENT
subtractors. Cint− and Cint+ then integrate ΔIo− and ΔIo+
ROUTING B US A RCHITECTURE
over a period T int that is determined by the time interval
In order to achieve high density, sensor arrays require a
between the negative edges of eval and sample (see Fig. 1(b))
small on-chip footprint for the measurement circuit. The area
to yield sensor output voltages V o− and V o+ . Here sample
of the CBCM sensor circuit can be significant because of the
is a variable delay pulse that allows for varying T int. If the
large device sizes necessary for improved matching and for
current pulses ΔIo± are approximated to be ideal pulses of
incorporation of additional calibration circuitry. In order to
amplitudes ΔIo± and widths T pw such that the total charge
realize differential sensor arrays, we have developed a current
delivered/removed to/from the integration capacitors remains
routing bus architecture. Fig. 2 illustrates the array architecture
the same, the sensor output voltages can be expressed as:
T int along side the pixel circuit and the timing diagram.
1 Each sensor pixel comprises six minimum size digital
V o± = · ΔIo± dt + V cm
Cint 0 transistors (M1-M4, M3’, M4’). In Fig. 1 nodes N− and N+
T int ΔCi connect the CBCM unit to the rest of the circuit, referred to as
≈ ±Ai · V step · · + V os± + V cm (3) the sensor evaluation module (SEM). In Fig. 2 nodes N− and
T pw Cint
where T int ≤ T pw, Cint = Cint+ = Cint− , and V os− N+ extend to form current bus lines that allow an entire array
and V os+ are the offset voltages from integration of Ios− of CBCM pixel units to share a common SEM comprising
and Ios+ . The differential output voltage ΔV o is given by: the complementary current mirrors, subtractors, integration
capacitors and output buffers. The timing diagram shown is
T int ΔCi for a row-wise select and a column-wise readout addressing
ΔV o = V o+ −V o− ≈ 2·Ai·V step· · +ΔV os (4)
T pw Cint scheme. Here row select, column select and sample are active
where ΔV os = V os+ − V os− . Under ideal conditions when low, and reset is active high. All pixels are reset globally in
both sides of the differential sensor are perfectly matched every clock cycle. In order to address a particular pixel for
ΔV os = 0. In reality, ΔV os = 0 due to device mismatch sensing, its corresponding row select first goes low followed
effects. From (4), T int being variable allows for gain tuning. by the column select enabling the sensor for evaluation. The
The relation between the sensor gain and T int is actually sampling of the SEM output is triggered by the negative edge
nonlinear because ΔIo− and ΔIo+ are transient current of the sample signal.
pulses and not ideal rectangular pulses as considered above. In a large sensor array implementing this architecture, the
The sample pulse also limits the sensor output voltage offsets bus-to-substrate capacitances of the current bus lines and
by limiting the durations over which Ios− and Ios+ are the source-to-bulk junction capacitances Csb ’s of transistors
integrated. The sensor output is finally buffered by a pair of M3 and M4 in the sensor pixel contribute to the parasitic
rail-to-rail buffer amplifiers B1 and B2 [11]. capacitances at the nodes N− and N+ . All these capacitances
166
column address decoder sensor pixel
N− column select N+ bus line
Csb Csb bus
address control and 3-phase clock generation line
M3 M4 Cbus-shield
row select
buffer
M3’ M4’ shield shield
reset line
row address decoder
M1 M2 (b) Cshield-sub
timing diagram
row select
substrate
reset
horizontal
column select cross section
of the
N− N+ sample
reset sensor evaluation module current bus
eval B4 Tint (c) (d)
sample B3
Vo− (a) Vo+
Fig. 2. (a) A fully differential capacitance sensor array based on a shielded current routing bus architecture. (b) Schematic of the sensor pixel. (c) Timing
diagram for pixel readout. (d) Horizontal view of the shielded current bus.
(a) (b) was simulated using Cadence Spectre for an input capacitance
1 2 range of ±20 fF. Input standing capacitances were assumed to
Differential Output Voltage ΔVo (V)
Tint = 195 ns
Tint = 60 ns be 10 fF. A sensing cycle period of 10 μs was used. The circuit
Vo+ employs 200 fF poly1-poly2 integration capacitors. The rail-
Sensor Outputs (V)
0.5 1
to-rail buffer amplifiers were designed to drive off-chip loads.
The buffer circuit offers an input capacitance of ∼ 100 fF. This
0 0 results in an effective integration capacitance Cint ∼ 300 fF.
Fig. 3(a) shows the transient response of the sensor after
−0.5 Vo_ −1 integration and buffering. ΔCi was varied between 0 fF
and 20 fF in steps of size 1 fF. Fig. 3(b) shows the static
calibration response curves obtained from transient simulations of the
−1 −2 curves extracted layout for T int = 60 ns and 195 ns. The mid-range
sensitivities of the sensor were estimated to be 90 mV/fF and
22 26 30 −20 −10 0 10 20
Time (μs) Input Capacitance ΔCi (fF) 130 mV/fF for T int = 60 ns and 195 ns, respectively, from the
plots. The figure also shows corresponding linear calibration
Fig. 3. (a) Transient response of the differential sensor for ΔCi between
0 and 20 fF. (b) Sensor static response curves with corresponding calibration curves computed for estimating the nonlinearity error of the
curves for T int = 60 ns and 195 ns. measurement circuit [3]. The slope of the calibration curve is
can sum up to 100’s of fF which degrades sensor evaluation the slope of the static response curve averaged over all the
speed. For this purpose a larger area metal shield fabricated simulation points. The nonlinearity errors were estimated to
in a lower metal layer can be used for isolating the bus line be 9.8% and 12.1% for T int = 60 ns and 195 ns respectively.
from the substrate as shown in Fig. 2. The effect of the bus-to- V. S ENSOR T ESTING
shield capacitance and the Csb ’s are canceled out by driving The sensor test chip comprised test structures with the
the shield line and the N-wells of transistors M3 and M4 in measurement circuit connected to metal3 (top most metal
each pixel with a potential that tracks the bus line potential. layer) plates configured according to the stray capacitance
This is achieved using the buffer amplifiers B3 and B4 as compensation scheme [9]. Fig. 4(a) shows a photomicrograph
shown in Fig. 2. This requires transistors M3 and M4 to be of one such test structure. The chip comprised 5 such test
placed in individual N-wells. Such shielding also improves structures with the metal3 plates varying in dimensions such
immunity of the current bus to substrate noise. In technologies that the metal3-to-substrate capacitances varied across the
with many metal layers the bus lines can be shielded from both input range of 0 to 20 fF. Fig. 4(b) shows a test structure
top and bottom. connected to interdigitated electrodes. This structure compen-
IV. S ENSOR S IMULATION sates for stray capacitances and overcompensates for standing
The sensor circuit was designed in a 0.5 μm CMOS capacitances such that the operation range of the measurement
technology for operation with a ±1.5 V supply. The circuit circuit lies in the linear portion of the static response curve.
was designed and laid out with a mirror gain Ai = 8. A base This can be useful for certain cell monitoring applications
transistor (M5 and M10) size of width 1.75 μm and length with critical linearity requirements [9]. The chip also included
1.75 μm was chosen for the design. The extracted layout 3-phase clock generation circuitry with the reset and eval
167
3
∆Ci = 19.01 fF
2
∆Ci = 11.45 fF
1.5
∆Ci = 7.62 fF
1 ∆Ci (fF)
∆Ci = 3.68 fF 3.2
noise (mV)
2.7
Measured
0.5 2.2
1.7
Fig. 4. Chip photomicrographs showing (a) a test structure with stray 1.2
capacitance compensation for measuring the standing capacitance of a metal3 0 0 5 10 15 20
168