A 1 V Low-Power Low-Noise DTMOS Based Class AB Opamp Herve F. Achigui1, Christian J.B. Fayomi2, Mohamad Sawan1 1
Polystim Neurotechnologies Laboratory, Electrical Engineering Dept. École Polytechnique de Montréal 2 Microelectronics Laboratory, Computer Science Dept. Université du Québec à Montréal E-mail :
[email protected]
Abstract — In this paper, we describe a novel class AB opamp based on dynamic threshold voltage transistors (DTMOS) for low voltage (1-V), low power and low noise applications. The opamp is used to build the front-end receiver part of a near infrared spectroreflectometry (NIRS) device. The opamp has a two-stage configuration; DTMOS pseudo pMOS differential input pairs are used for input common-mode range enhancement, followed by a single ended class AB output. Experimental measurements from previous designs confirm the usage of a DTMOS device to build a 1-V opamp, using standard 0.18-µm CMOS technology. The performed post-layout simulation results show an input referred noise of 107 nV/√Hz at 1 kHz, and a power consumption of 33.1 µW under 5 pF and 10 kΩ loads. The dc open loop gain is 60 dB, and a unity frequency of 2.73 MHz. The opamp has a CMRR of 100 dB, and input and output swings of 0.6 V and 0.8 V respectively.
I.
INTRODUCTION
O
PTICAL TOMOGRAPHY is a new and widely expanding medical imaging technique that uses near infrared (NIR) light as the probing radiation. Particularly in medicine and biology, this is becoming one of the most important topics, providing researchers with means of using NIR light as a probe for tissue examination and inspection. One of the applications of such a system would be to monitor epileptic seizure precursor signals in a patient. Epilepsy is generally defined as a neurological disorder characterized by sudden, recurring attacks of sensory or psychic malfunction with or without loss of consciousness or convulsive seizures. Seizures result from abnormal electrical activity in the brain, and can be generalized or partial. Before each seizure, the body uses remarkable amounts of energy to sustain the seizure; the danger here is that the brain is using up available oxygen very rapidly. This has been reported in [1] as energy bursts measured several hours before clinical onset using EEG. The monitoring of cortical tissues’ oxygen variation would enable us to identify such a phenomenon. Progress in medical imaging technology is driven by the desire to obtain earlier and more accurate assessment of disease occurrence, disease course and efficacy of treatment. NIR light offers the advantage of having much lower absorption than visible light among existing imaging systems. Nearinfrared spectroreflectometry (NIRS) is the usage of light in the NIR range to determine cerebral oxygenation, blood flow, and metabolic status of the cortical tissues, based on the measurement of the variation of tissue optical properties such as
0-7803-8935-2/05/$20.00 ©2005 IEEE.
absorption, refraction, anisotropy and scatter. It provides a portable, noninvasive means of monitoring and imaging brain function and biological tissues, because of the relatively low absorption of water and high absorption of oxy– and deoxyhemoglobin in the NIR region of 600–900 nm. Due to these properties, the NIR light can penetrate biological tissues in the range of 0.5–3 cm, providing the ability to investigate deep tissues, and differentiate between healthy and diseased tissues. The proposed NIRS device is composed of two parts: the emitter and the receivers. The emitter consists of three NIR laser diodes emitting light at discrete wavelengths of 735 nm, 840nm and 940 nm. The receivers are a set of six separate identical detectors, optically and electrically isolated from each other as depicted in Fig. 1 (a). Each detector consists of a CMOS photodiode with a built-in current-to-voltage converter, voltage opamp, filter, and mixer as shown in Fig. 1 (c). The transmission of light emitted by the NIR laser diodes that travel through cortical tissues is dependent on the reflectance, scattering and absorption that occur in each particular tissue. A fraction of the injected light photons survive to return and exit the skin’s surface after being strongly scattered inside the scalp, skull and brain, following the “banana” shaped path illustrated in Fig. 1 (b). The amount of reflected photons depends on the power of the light source and the source-to-receiver distance. For a five layer head model (scalp, skull, cerebrospinal fluid (CSF) layer, gray and white matters), the attenuation level with a source-to-receiver distance of 4 cm can be approximated by equation (1) as reported in [2], I out ∈ [5, 12] × 10−8 I in
(1)
where Iin is the incident light intensity and Iout the intensity of the detected light. In this paper, a new class AB opamp for the NIRS front-end receiver of a wireless brain oxymeter is proposed. The circuit makes use of DTMOS folded cascode differential input pairs to increase the input common mode range (ICMR), as well as to enhance low noise, low power and low voltage (1-V) operation. Section II focuses on the description of the NIRS front-end receiver. Preliminary experimental results from the design in [3] are presented in section III, followed by the obtained post layout simulation results.
(a)
(b)
information. Moreover, the useful voltage signal will still be embedded in additional noise generated by various mechanisms such as a non-uniform optical path length, variation of the incident light angle, and the transistors’ electronic noises. The opamp is the most critical part in noise performance for the proposed NIRS instrument. The opamp characteristics (dc gain, dominant and second pole frequency, input and output resistance, and positive and negative slew rate) are chosen to significantly match the operation performance of the DTMOS amplifier, as reported in [3], and are optimized for overall system performance as are the power consumption, noise and ability to amplify microvolt signal. III.
(c) Fig. 1: NIRS emitter–receiver system: (a) NIRS module, (b) Light propagation path, (c) Block diagram of one channel of the NIRS front-end receiver.
II.
DESCRIPTION AND MODELING OF THE NIRS FRONT-END RECEIVER
The NIR lights are modulated by a stabilized sinusoidal current at a frequency between 10 and 50 kHz. The sources are placed on the scalp of the subject, and are activated one at a time. The path of the light photons from each source that propagates through the cortical tissues and is reflected to the surface of a single detector is described as a channel. The light source in each channel is modeled by a sinusoidal current signal. The intensity of the reflected light is dependent on the optical path length that the detected light travels within each homogenous region (scalp, skull, CFS layer, grey and white matter). Thus, the amount of captured photons at the surface of each detector will vary from one detector to another within each hexagonal patch element. The CMOS photodiode transforms the reflected light into current. A transimpedance is used to transform the currents extracted from the photodiodes into voltages, and the low noise, low power, DTMOS-based class AB opamp is used to selectively amplify the low amplitude signal before it is filtered, and then demodulated, as described in the block diagram of Fig. 1 (c). The Verilog-A model for the class AB opamp has been developed and implemented using spectreS under Cadence. Since the thickness and the structure of superficial tissues such as the scalp and the skull vary from one individual to another, they affect the NIR measurements. Therefore, in each channel and for each layer, the optical path length of the photons traveling through the head varies, resulting in a variation of the intensity of the reflected light on the surface of the detector module. A Gaussian distribution function is used to model this variation in amplitude attenuation level in each channel. According to equation (1), the reflected light intensity, which is modeled as photodiode current, would be in the order of a few nano-amperes (nA). The gain of the transimpedance (TIA) module is set to approximately 80 dB(Ω). This block provides the following stages with a voltage signal amplitude of a few microvolts. Most of the current noise signal will still pass to the following stages, and is amplified along with useful signal
OPAMP DESIGN AND RESULTS
A DTMOS device is a transistor whose gate is tied to its substrate. Consequently, the substrate voltage in DTMOS changes with the gate input voltage, and causes the threshold voltage (Vth) to change accordingly. As a result, there is no floating body effect, and the Vth swing is maximized, resulting in an improvement of the Ion/Ioff ratio. The proposed DTMOS-based opamp input stage uses a folded cascode wide swing current mirror as presented in Fig. 2 (b), [3]. However, the circuit characteristics have been optimized, and make it suitable for biomedical applications which require low voltage, low power and low noise analog circuits. Input devices, pMOS based DTMOS M1 and M2, are operated between weak and moderate inversions to ensure transconductance efficiency (gm/ID) for the minimum input referred gate noise voltage at the minimum possible bias current. MP1, MP2, and MP3, are DTMOS transistors which provide the required bias current to differential input and folded cascode transistors. In addition, operational transconductance amplifiers (OTA) have been used to provide adequate bias voltages to MN3 and MN4 as depicted in Fig. 2 (b), and for enhanced small signal operation stability. The operation of the class AB output stage was presented in detailed in our previous work. A. Measurement results from our previous design This is part of our continuous work to prove that the usage of pseudo pMOS based DTMOS devices provide many advantages in standard CMOS technology, and can be widely used to build low voltage (< 1 V) circuits. The chip of the design proposed in [3] was fabricated in a 0.18-µm single poly, six metal salicide CMOS process. The die micrograph is shown in Fig. 3. Fig. 4 (a) shows the measured opamp response to a 0.6 Vpp sinusoid input signal, and Fig. 4 (b) depicts the step input response under a 10 KΩ and 5 pF load. Measurements were conducted using the Agilent 33250A function waveform generator and the Tektronix TDS7154 oscilloscope. The later design exhibits large power consumption (~ 600 µW), which makes it unsuitable for biomedical handheld devices such as the wireless NIRS brain oxymeter apparatus we are building. B. Design optimization for low-noise and low-power Equation (1) outlines the need for a receiver which is capable of processing ultra low-amplitude light signals. Amplification
(a)
(b)
(a)
(c) Fig. 2: Schematic of our current DTMOS-opamp: (a) block scheme, (b) differential input stage, (c) class AB output stage.
(b) Fig. 4: Measured transient results from our first opamp design: (a) 0.6 Vpp sine response, (b) step input response. Fig. 3: Photomicrograph of our first opamp.
of such low level signals imposes severe noise conditions. On the other hand, the major concern in conventional analog CMOS-based circuits operating at low frequency is the flicker noise (1/f), which is very harmful in low-frequency applications such as the present one, because of its power spectrum and voltage offset. Noise analysis of the circuit in Fig. 2 (b) is calculated by taking into account only transistors in the signal path (M1, M2, MN1, MN2, MP2 and MP3), and is given by general equation (2). WM 2 , LM N 2 and LM P 3 are independent parameters; increasing either will decrease the 1/f noise. After choosing LM
N2
and
LM P 3 for best thermal noise, the minimum achievable noise is
obtained by choosing LM 2 according to equation (3): vni2 =
Kf p Δf
μ p C WM LM 2 ox
2
2
∂vni2 =0 → ∂LM 2
⎡ Kf ⎢1 + n f ⎢ Kf p ⎣
⎛ LM 2 ⎜ ⎜ LM ⎝ N2
2
⎞ ⎛ LM 2 ⎟ +⎜ ⎟ ⎜ LM ⎠ ⎝ P3
Kf p 1 1 1 =2 + 2 Kf n LM N 2 LM P 3 LM 2
⎞ ⎟ ⎟ ⎠
2
⎤ ⎥ ⎥ ⎦
(2)
(3)
where Kf is the flicker noise coefficient, a process dependent parameter, µ is the electron (or hole) mobility for NMOS
(PMOS), Cox is the gate capacitance per unit area, W and L are the gate width and length, and ∆f is the bandwidth at frequency f. Since the entire system is expected to be made of a large number of detector modules (fifty-six), the power consumption for each detector must also be minimized. DTMOS devices are used for ICMR enhancement, resulting in the maximum achievable input dynamic range (DR) for this circuit topology, while preserving the minimum power consumption [3], [4]. The input devices M1 and M2 have a W/L ratio of 2733 to ensure subthreshold voltage operation under a bias current of 4.23µA per device. Also, due to the current path from the gate to the substrate, the input impedance of a pMOS based DTMOS opamp would be lower in comparison to an opamp with standard differential pMOS input devices. This outlines the need for the class AB output stage to provide current to drive resistive loads, whereas the preamplifier proposed in [4] is suitable for driving capacitances load, and Authors use a buffer to isolate each stage from the others.
Fig. 5: Layout of our current DTMOS based class AB opamp.
C. Post layout simulation results from our current design The circuit schematic was implemented and simulated with Cadence using Hspice. The post layout simulation of the extracted circuit was conducted using spectreS in a 0.18 µm CMOS standard technology, under a 10 KΩ and 5 pF load. The common centroid technique has been used to minimize the effect of threshold and process mismatches on differential input pairs and on current mirrors for the opamp, as depicted on the layout of Fig. 5.
Fig. 6: Current DTMOS based class AB input referred noise voltage.
The obtained results from the post-layout simulations of the opamp presented in Fig. 6 give an input referred noise of 107 nV/√Hz at 1 kHz. The power consumption is only 33.1 µW. Fig. 7 illustrates the open loop dc voltage gain of 60 dB, the phase margin of 62 degrees and a GBW of 2.73 MHz. The opamp has a common mode rejection ratio (CMRR) of 100 dB, and input and output swings of 0.6 V and 0.8 V respectively. IV.
CONCLUSION
A versatile DTMOS-based class AB opamp for 1 V applications has been proposed. The circuit is implemented to build the analog front-end part of a NIRS receiver of a multiwavelength wireless brain oxymeter apparatus. The entire NIRS receiver is capable of operating with supply voltage as low as 1 V. The post-layout simulation for the opamp gives an input referred noise of 107 nV/√Hz at 1 kHz, and a power consumption of 33.1 µW. Applications for the proposed system should extend from a wireless multi-wavelength brain oxymeter to a real-time imaging handheld device of brain activity for patients suffering from mental diseases, such as epilepsy and Parkinson’s.
[1]
ACKNOWLEDGMENTS
[2]
The authors would like to acknowledge the financial support of the Natural Sciences and Engineering Research Council of Canada (NSERC), the Microelectronics Strategic Alliance of Quebec (ReSMiQ) and the Canadian Microelectronics Corporation (CMC).
[3]
Fig. 7: Post-layout frequency results from our current DTMOS based class AB opamp.
REFERENCES
[4]
B. Litt, et al, “Epileptic Seizures May Begin Hours in Advance of Clinical Onset : A Report of Five Patients,” Neuron, vol. 30, pp. 51–64, April 2001. E. Okada, D.T. Delpy, “Near-infrared light propagation in an adult head model. II. Effect of superficial tissue thickness on the sensitivity of the near-infrared spectroscopy signal,” Applied Optics, vol. 42 (16), pp. 2915-22, June 2003. H.F. Achigui, C.J.B. Fayomi, M. Sawan “A DTMOS-based 1 V opamp,” Proc. IEEE ICECS, vol. 1, pp. 252–255, Dec. 2003. H.F. Achigui, M. Sawan, C.J.B. Fayomi, “A 1 V Low Power, Low Noise DTMOS based NIRS Front-End Receiver,” in Proc. WMSCI, Jully 2005.
