Measuring and Extraction of Biological Information on New Handheld Biochip-Based Microsystem

1

Measuring and Extracting BiologicalInformation on a new Hand-held

Biochip-based Microsystem

P. A. C. Lopes, J. Germano, T. M. Almeida, L. Sousa, M. S. Piedade,

F. Cardoso, H. A. Ferreira and P. P. Freitas

Abstract

This paper presents the techniques developed for the extraction of biological information

in a recently developed hand-held biochip-based microsystem. The microsystem is based

on a magneto-resistive array biochip composed of a number of sensing sites with magnetic

tunneling junctions (MTJ) and diodes. To drive the MTJ, different techniques are addressed

with different types of signals. Different filtering strategies are also studied, which allow the

recovery of bio signals from the noise without increasing too much nor the time required to

access all the sensors, nor the power consumption of the board. In conclusion, experiments

with the system in a setup to detect actual bio signals are presented with encouraging results.

Index Terms

biochip, microsystem, magnetic sensors, bio-molecules, signal processing.

I. Introduction

One of the trends of the last decade has been the miniaturization of typical large lab-

oratory experiments. This was made possible by the advances in microfluids and Micro-

Electro-Mechanical Systems (MEMS) technologies. One of the outcomes of this trend has

been the so called “lab on a chip” systems [1]. For lower scale production, microsystems such

P. A. C. Lopes and others are with Instituto Engenharia de Sistemas e Computadores, Investigao e Desenvolvimento(INESC-ID) and Instituto Superior Tcnico, INESC-ID, Rua Alves Redol n 9, 1000-029 Lisboa, Portugal, phone:+351-213100379, email: [email protected]

F. Cardoso and others are with Instituto Engenharia de Sistemas e Computadores, Microsistemas e Nanotecnologias(INESC-NM), email: [email protected].

July 7, 2006—4 : 31 pm DRAFT

MEAS. AND EXTRAC. OF BIO. INFO. ON HANDHELD BIOCHIP-BASED MICROSYS., IMTC 2006 PAPER #66262

as the one described in this paper offer great promise. The microsystem described in this

paper [2], is based on magnetoresistive biochips. These chips have been introduced for fully

integrated biomolecular recognition assays [3, 4]. In these experiments, target bio-molecules

are marked with magnetic particles and are subsequently recognized by bio-molecular probes

immobilized at the surface of the chip over sensing sites. The markers fringe fields are then

detected by magnetic tunneling junctions (MTJs) [1].

The developed system consists of a compact, credit card size, portable hand-held mi-

crosystem for biomolecular recognition assays.

II. Architecture

The proposed architecture for the biochip platform is organized in two main modules

(fig. 1): i) the Sensing and Processing Module (SPM) and ii) the Fluid Control and Com-

munications Module (FCCM). The SPM integrates the biochip and provides the circuits

that directly interact with the array of biosensors (biochip). The FCCM interfaces the

platform with the external world by controlling the fluid carrying the magnetically tagged

biomolecules and by providing wireless communication with a handheld analyzer based on a

Personal Digital Assistant (PDA).

MCSPI

Biochip Platform

Wireless

USBHeater control

MC/DSP

TemperatureControl

Current line control

PDA / Laptop

Emitter/Receiver

Software Interface

DriverFluid Flux Control

Emitter/Receiver

Bluetooth Module

Row address

Sensing and Processing Module (SPM)

Fluid Control and Communication Module

(FCCM)

Driver

Magnetic Field Gen. SPI

VSensor

VRef

Current Gen.

Fluid Source

Waste Container

Pump and Valves

Biochip

...

Sw

itch

SPI

Co

ntro

l

RS

232

Amplified signal

Signal Acquisition

Measure type

SPISensor

Addressingand Readout

Column addressSwitch

...

VFeed

Fig. 1. Full diagram of the microsystem.

July 7, 2006—4 : 31 pm DRAFT


A. Current generator circuits

To perform readings in the complete sensor array, the current to the sensor is gener-

ated using a Digital to Analog Converter (DAC) and a voltage-to-current converter and is

multiplexed into the biochip. Figure 2 depicts the circuit diagram employed in the cur-

rent generator. The presented circuit guaranties that the current that flows to the sensor

Voltage to current

conversion

Avdd=5V

Current mirror for reference

sensor

Re Re

+

-QFeed

R1

Qm QRef

Im IRef

IFeed

I≈0

Current mirror for

sensor

Re

QSensor

ISensor

Sensor and multiplexing

RFeed

10-bit DAC (C)

SPI

VRef

Fig. 2. Current generation circuit.

is the same that runs through the Operational Amplifier (OPAMP) feedback resistor, thus

eliminating the temperature and the current errors introduced by the mirror and by the

transistor.

For the magnetic field generator, a circuit similar to the one used for the generation of

the sensor drive current was used. The current intensity in the coil, and consequently the

magnetic field, is controlled using the DAC, and scaled through a resistor.

III. Sensing sites

The biochip is composed of a number of sensing sites, formed by a MTJ and a diode. The

sites are arranged in a array and accessed through line and column wires that are selected

through multiplexers outside the chip. The diode has two functions: to act as a commutator

that prevents access to sites (through loops), other than the selected one, and to act as a

temperature sensor for biological reactions that take place on the chip. A typical set of

magnetic tunnel junctions used in the biochip were characterized in [5]. The resistance of

July 7, 2006—4 : 31 pm DRAFT


the MTJ varies with the transversal component of the applied magnetic field. An important

characteristic of the junction is its tunneling magnetoresistance ratio, TMR. This is given

by,

TMR =Rmax −Rmin

Rmin

(1)

where Rmax and Rmin are, respectively, the maximum and minimum resistance values ob-

tained with magnetic opposite saturation fields. The resistance variation follows a hysteresis

curve. Assuming that the junction is being driven by a current I0, then the sensitivity to

the magnetic field of the measured voltage signal is given by,

∂v

∂h= Sv

H = TMR(V)RJ

∆Hmax

I0. (2)

IV. Excitation and Acquisition

In order to measure the resistance of the MTJ, a know current was applied at the site

and the resulting voltage was measured.

A. AC and DC measures

Applying a current through the sensing site and measuring the voltage signal, will result

in a signal that combines the voltage drop across the diode and across the MTJ. In order to

extract the diode signal, one can simply subtract the signal measured before the insertion of

the particle solution at the sensor. This was the procedure used in fig. 6. This will result is

a signal that is proportional to the concentration of particles in the solution.

However this is a small signal, that is embedded in a large signal. These signal levels

requires the AD converter to have a large dynamic range. This problem can be reduced if

the applied external magnetic field has a sinusoidal component. This alternating magnetic

field will produce a corresponding variation in the resistance value. When this is multiplied

by the values of the current source, it will result in an AC signal, with the same frequency

July 7, 2006—4 : 31 pm DRAFT


as the one used to generate the magnetic field. This signal can then be separated from the

DC voltage through a high pass filter, and then amplified.

B. Selecting the excitation current value

The TMR of the junction decreases with applied voltage, being maximum with zero

applied voltage. For the measured set used in [5] the TMR was kept constant until about

30mV. At this level, a TMR of about 27% was obtained. The TMR then decreases almost

linearly with the applied voltage, in a range up to 500mV. This can be modelled as,

TMR(V)

TMR(0)= 1− V

2 V1/2

(3)

where V1/2 is about 350mV. This suggests the use of low polarization voltages. However, a

low polarization voltage implies a low drive current. The sensitivity of the measured voltage

signal to the magnetic field is given by (2). For low currents, the TMR is high, but since

the current is low, the sensitivity will be low. For high currents the TMR is low, making

the sensitivity low. The current value that maximizes the sensitivity lies somewhere in the

middle. This can be calculated using the referred equations, resulting in,

I0 =V1/2

RJ

. (4)

For the junction in [5], which has about 14.4kΩ, the optimum current is at about 30µA.

Note that the drive current was optimized for the highest signal level. If the requirements

were to optimize the signal to noise ratio, then the results would be different. However, for

this application the noise level is not limited by the junction.

Due to the ultra low thickness of the dielectric, the MTJ may breakdown for applied

voltages over 1.1V. For example, for MTJ’s with RJ = 15.3kΩ the maximum drive secure

driving current is 65µA.

July 7, 2006—4 : 31 pm DRAFT


0 0.05 0.1 0.15 0.2−120

−100

−80

−60

−40

−20

0

Frequency (kHz)

Pow

er/F

requ

ency

(V

dB/H

z)

Noise Power spectral density

Fig. 3. Noise power spectral density at the AD converter, with the DA configured to a maximum currentof 1mA

V. noise

Measurements of noise levels in the board were made, for the case of a sinusoidal current

excitation signal (AC mode). The sampling frequency was of 480Hz. The signal was a

30Hz, 5µA current, injected through a 10kΩ resistance. The noise power spectral density

can be seen in fig. 3. The noise is mostly composed of four components: Harmonics of

the 30Hz frequency, quantization noise from the DA converter; 50Hz power line frequency

noise; low frequency noise; and white noise. The total noise level is about 1mVRMS, due to

DA quantization. The 50Hz power line noise and low frequency noise amount to 370µVRMS

and the resulting noise is about 37µVRMS. Further filtering with a 3.3s length band-pass

filter, reduces the noise to about 8µVRMS. The white noise floor at about −100dBV/√

Hz =

10uV/√

Hz is mainly due to the noise figures of the DA.

The high noise floor and quantization noise lead to a change of the DA scale from a

maximum of 1mA to only 100µA. This not only reduces the quantization noise but also the

white noise floor. The noise floor was reduced by a factor of ten, to a reasonable 1uV/√

Hz,

as shown in fig. 4. This figure represents the power spectral density of the noise with

measurement done in AC mode, and with the amplifier set for a gain of ten. Also shown is

the noise level when the load is set to a sensing site diode, where a high low-frequency noise

July 7, 2006—4 : 31 pm DRAFT


is visible, which lead to an increase of the signal frequency to 325Hz, with a sample ratio of

8spl/periode and a conversion ratio of 2600 cvs/s. The anti-aliasing filter of the AD had a

3dB cut frequency of 1696Hz, and an effective noise bandwidth of 3148Hz. In this figure

there is a strong 50Hz interference signal, but it can be removed digitally.

0 500 1000 1500−140

−120

−100

−80

−60

−40

frequency (Hz)

Pow

er/F

requ

ency

(V2 d

B/H

z)Noise Power spectral density

33KOhm Resistorsensor site PIN diodo

Fig. 4. Noise power spectral density measured by the AD converter, with the system configured to AC modewith gain 10, and the DA configured to a maximum current of 100µA. The peak at 325 Hz is the appliedsignal.

Finally fig. 5 presents the PSD of the noise at the input of the AD converter under the

same conditions as in fig. 4, but measured with a spectral analyzer. The noise levels are

in close agreement, and it can be seen that the simple RC anti-aliasing filter at the input

of the Sigma/Delta AD converter is effective in removing the high frequency noise. Signal

harmonics due to the sample-and-hold at the output of the DAC are highly visible.

VI. Signal Processing

In the system there are two main tasks for the signal processor embedded in the board:

the generation of the excitation signal and the recovery of the biological signal. Some of the

signal processing techniques rely on previously obtained [2,6] models for the MTJ and diode.

July 7, 2006—4 : 31 pm DRAFT


100

102

104

106

10−8

10−6

10−4

10−2

100

Noise Power Spectral Density

RM

S F

requ

ency

−1/

2 (V

dB

Hz−

1/2 )

frequency (Hz)

Fig. 5. Signal and noise levels at the input of the AD converter, made with a spectral analyzer, with thesystem set to AC mode with gain 10. The chart is a superimposition of three different measurements withdifferent measurement bandwidths. The large bandwidth of the last measurement filters out the peaks seenat lower frequencies. The peak at 325 Hz is the signal, following peaks at signal harmonics..

A. Generating the excitation signal

For AC measurements, the chosen excitation signal was sinusoidal, while for DC a con-

tinuous current was used. The sampling ratio was chosen to be a constant multiple of the

sinusoidal signal. This allowed the signal to be generated very simply through a lookup

table.

B. Noise Filtering

Assuming AC excitation, either with an AC field or an AC current, the amplitude of

the voltage signal at the sensing site must be determined. This can be done using several

techniques. The measured signal, y[n], can be approximated by sinusoidal signal, with

analog frequency f0, which correspond to the digital frequency w0 = 2πf0/fA. The signal is

corrupted by white noise, v[n], with standard deviation σv,

y[n] = A cos(ω0 n + φ) + v[n], (5)

and the goal is to estimate the amplitude, A, of the signal. A number of techniques are

described. The DFT amplitude estimator was chosen for implementation.

July 7, 2006—4 : 31 pm DRAFT


B.1 Direct RMS value calculation

The first approach is simply to calculate the RMS value of the received signal, as given

by (6). This produces reasonable results as long as the noise level is low. Assuming that N

samples are taken from the received signal, where N is a multiple of the signal period, then

the signal amplitude estimate will be,

ARMS =

√√√√ 2

N

n0+N−1∑n=n0

y[n]2. (6)

This amplitude estimator is biased, its expected value is given by

E[ARMS/A] = 1 + (σv/A)2, (7)

and the standard deviation, or RMS value, of the noise will be, σ(ARMS/A) =√

2N

σ(v/A). In

our application the error will usually be dominated by the bias, since the value of N can be

large.

B.2 DFT amplitude estimator

Since we intend to calculate the amplitude of the received signal, an obvious approach

is to calculate its DFT, and determine the amplitude at the excitation frequency. However,

it is not required to calculate the full DFT, only its amplitude at the given frequency. This

can be calculated by,

ADFT real =

∑n0+N−1n=n0

2y[n]cos(ω0 n)

N(8)

ADFT imag =

∑n0+N−1n=n0

2y[n]sin(ω0 n)

N(9)

ADFT =√

A2DFT real + A2

DFT real (10)

This is an unbiased estimator of the amplitude. Its standard deviation, for small values of

the noise signal, is given by,

σ(ADFT/A) =

√2

Nσ(v/A). (11)

July 7, 2006—4 : 31 pm DRAFT


This result is the same as the one from the RMS calculation. However, since this estimator

is unbiased the resulting estimation error is usually much lower.

B.3 Optimal Passband filter

Assuming a model for the noise signal in (5), namely its power spectral density or

its autocorrelation function, an optimal filter to remove the noise can be determined, in a

form of a FIR filter [7] with impulse response wj and length N . This can be formulated

as a winner filtering problem [8], where, d[n] = A cos(ω0 n + φ) in (5) is the desired signal

and y[n] is the input signal. The input y[n], can be decomposed in two components, y0[n] =

Acos(ω0 n+φ) and v[n]. Only v[n] is considered stochastic. If one assumes the noise is white,

as is approximately the case around the excitation frequency (section V), then the optimum

filter results in the truncation of a sinusoidal signal. Defining y0 = [y0[n], ...,y0[n−N + 1]]T ,

and w = [w0, ...,wN−1]T , the autocorrelation matrix of the signal is given by, R[n] = y0y0

T +

σ2v δ[i− j] and the crosscorrelation vector is P[n] = y0 y0. This will result in a time varying

optimal filter W [n]. The output of this filter is then sampled at its maximum to determine

the amplitude of the sinusoid, resulting in,

w =y0 y0[nmax]

y0Ty0 + σ2

v

. (12)

For the given y0, as long as σ2v is low, and N is a multiple of the period, this results in the

estimator,

AOpt =

∑n0+N−1n=n0

2 y[n] cos(ω0 n + φ)

N. (13)

This is also an unbiased estimator and the standard deviation is the same as for the DFT

amplitude estimator. However, this technique requires the knowledge of the phase of the

measured signal, so the previous technique was used.

July 7, 2006—4 : 31 pm DRAFT


VII. Experiments

The microsystem was tested using a solution of 2.3× 109particles/ml with 1.5µm diam-

eter magnetic nanoparticles. An 5µA DC current was driven by the DAC through a 10kΩ

MTJ. The voltage signal was measured by an AD converter at a sample rate of 6 Hz after

passing through a suitable anti-aliasing filter. The measurement time was about 8 minutes.

The measured signal is presented in fig. 6, after the removal of a 47mV DC signal. The

solution was dropped on the sensor after about 1000 samples, and after about 1750 samples,

the sensor was washed with distilled water. The figure clearly shows a 150µV signal due to

the presence of nanoparticles, demonstrating that the microsystem can be used for particle

detection.

Idc = 5 µA

Hdc= 15 Oe

0 500 1000 1500 2000 2500 3000

-50

0

50

100

150

200 wash wit h DI water

Number of samples (6 spl / s)

10 µL ofpart icles(1.5 µM)

∆ µVV = 191

∆V (

µV)

Fig. 6. Time variation of the measured signal for evaluation of particle detection capabilities.

VIII. Conclusion

Techniques to measure and extract biological information in a recently developed hand-

held biochip-based Microsystems were presented. Different types of measures were compared.

Also studied were different filtering strategies based on noise signals presented in the sys-

tem. These strategies allowed the filtering of the noise without excessively increasing the

total time required to measure the signals at the sensors, while maintaining low computa-

tional complexity and low power consumption of the board. Finally, the viability of the

July 7, 2006—4 : 31 pm DRAFT


system was demonstrated by testing it in a real bio signal detection experiment with good

results.

References

[1] D. M. W. Shen, X. Liu and G. Xiao, “In situ detection of single micron-sized magnetic

beads using magnetic tunnel junction sensors,” Appl. Phys. Lett., vol. 86, no. 25, p.

253901, June 2005.

[2] M. S. P. et al., “Microsystem for biological analysis based on magnetoresistive sensing,”

in Instrumentation and Measurement Technology Conference 2006, Sorrento, Italy, 2006.

[3] M. Johnson, Magnetoelectronics. Academic Press, 2004, ch. 7, pp. 331–274.

[4] H. A. F. D. L. Graham and P. P. Freitas, “Magnetoresistive-based biosensors and

biochips,” Trends Biotechnol, vol. 22, no. 9, pp. 455–462, September 2004.

[5] T. M. A. et al., “Characterisation and modelling of a magnetic biosensor,” in Instrumen-

tation and Measurement Technology Conference 2006, Sorrento, Italy, 2006.

[6] ——, “Magnetoresistive biosensor modelling for biomolecular recognition,” in XVIII In-

ternational Measurement Confederation World Congress, Rio de Janeiro, Brazil, 2006.

[7] A. V. Oppenheim and R. W. Schafer, Discrete.Time Signal Processing. Prentice Hall,

1999.

[8] S. Haykin, Adaptive Filter Theory. Prentice-Hall, Inc., 1996.

July 7, 2006—4 : 31 pm DRAFT

Measuring and Extraction of Biological Information on New Handheld Biochip-Based Microsystem

Documents

Transcript of Measuring and Extraction of Biological Information on New Handheld Biochip-Based Microsystem