A Low-Noise CMOS Receiver Frontend for MRI Jens Anders 1 and Giovanni Boero 1 1 Ecole Polytechnique F´ ed´ erale de Lausanne Lausanne, Switzerland Email: jens.anders@epﬂ.ch Abstract—In this paper a novel architecture for an integrated receiver front-end for micro magnetic resonance imaging (micro- MRI) applications is described. While the chip consumes only 9 mA supply current (4 mA in the LNA and 5 mA in the output buffer) from a 3.3V power supply, it has a measured input referred noise density of only 0.6 nV/ √ Hz . The receiver consists of a reception coil, an on-chip tuning capacitor, a low-noise ampliﬁer, and a 50 Ω output buffer. The system is designed for operation in a B0-ﬁeld of 7 T corresponding to a frequency of 300 MHz. It is implemented in a 0.35 μm CMOS high-voltage process and occupies a chip area of 850 μm × 500 μm. I. I NTRODUCTION Over the last three decades, magnetic resonance imaging (MRI) has become one of the most important medical imaging techniques. Its major advantage lie in the fact that it offers a much greater soft tissue contrast than computed tomography (CT) and the ease with which detailed images in any body plane can be obtained compared with other imaging tech- niques. Since its introduction in 1973, signiﬁcant progress has been made in terms of contrast, spatial resolution and scanning time. While there has been considerable effort in the design of microcoils for very small sample sizes, e.g. [3], surprisingly few attempts have been made to integrate these microcoils with on-chip electronics. In this work, we present a microcoil with a diameter of 500 μm which has been cointegrated with a low-noise ampliﬁer and a 50 Ω buffer in a conventional CMOS process. Before discussing the details of the design, a short review of the working principles of MRI will be given in section II that introduces the standard MRI terminology and summarizes the various constraints that require tradeoffs in the design of an integrated receiver for this speciﬁc application. Section III discusses the receiver circuitry in detail and the corresponding measurement results are discussed in section IV. The paper closes with an outlook on future work and some concluding remarks. II. MRI WORKING PRINCIPLE AND DESIGN REQUIREMENTS A. Working Principle The physical mechanism underlying MRI is nuclear mag- netic resonance (NMR). NMR exploits the fact that nuclei possessing an odd number of protons or neutrons have an intrinsic magnetic moment. For a spin-half particle in an applied static magnetic ﬁeld (B 0 -ﬁeld) there are only two observable energy states. The energy difference between these states, ΔE, depends on the B 0 -ﬁeld strength as well as on the gyromagnetic ratio of the nucleus, γ , according to ΔE =¯ hγB 0 . (1) If now an additional RF-magnetic ﬁeld (B 1 -ﬁeld) is applied to the system, resonant absorption will occur if the frequency ν of this ﬁeld is close to the resonant condition ν ≈ ν 0 = ΔE h = γB 0 2π , (2) where ν 0 is the so-called Larmor frequency. In a pulsed-NMR experiment such as those used in MRI, it is the relaxation of the system after the B 1 -ﬁeld has been switched-off which is observed. To make use of the NMR phenomenon for imaging, the Larmor frequency of the nuclei is locally changed by gradients in the B 0 -ﬁeld in all three spatial dimensions. Thus, the infor- mation about local spin densities is encoded in the frequency domain 1 and can be retrieved by means of an inverse Fourier- transform. B. Design Constraints The main design goal for an MRI receiver is to provide the highest possible SNR at its output for all operating conditions. Fortunately, the operating conditions are generally predictable in an MRI environment, i.e. there is a clear upper limit on the induced voltage and there are no strong interferers present. Therefore, the most critical design goal for an MRI receiver front-end is superior noise performance while the distortion properties are of less importance as the receiver will always work in its linear operation region. To achieve this goal, both the reception coil and the LNA have to be optimized for SNR. For the coil this translates into geometry optimization and for the LNA into design for optimal noise ﬁgure. Since the most reasonable choice for the LNA is a voltage ampliﬁer, and induced gate noise can be ignored for the conditions encountered in MRI receivers, which is due to the fact that the coil resistance is smaller than the optimal source resistance for noise matching, the input transistors’ drain noise is the dominant noise contributor [11]. Therefore, optimizing the noise ﬁgure is equivalent to minimizing the input referred 1 In practical applications, gradient sequences also encode information in the phase domain. 978-1-4244-2879-3/08/$25.00 ©2008 IEEE 165