The proposed research focusses on the issues associated with 1) the development of microelectromechanical system (MEMS) active microphones; 2) the integration of the MEMS with CMOS aVLSI electronics to provide adaptive feedback; 3) appropriate coding of sound for interpretation; and 4) the testing of the microphone in realistic situations. The project has the following aim:
To build and test an active multiple channel MEMS/CMOS microphone with independent feedback control for each channel, operating over a wide range of frequencies, generating an output suitable for interpretation of the sound field.
To meet the above aim, the following objectives have been identified:
• To design and fabricate a single active MEMS microphone
• To design and fabricate an active MEMS-array microphone
• To design MEMS/CMOS interface aVLSI with adaptive feedback control
• To integrate an active MEMS/CMOS microphone with adaptive feedback control
• To characterise the active MEMS/CMOS microphone output
• To test the microphone with a range of acoustic stimuli in a range of acoustic environments
• To assess the interpretation of the output from this microphone.
The application of adaptive micro-electro-mechanical systems (MEMS) device in biologically-inspired cochlear model (cochlear biomodel) is seen as a preferable approach to mimic closely the human cochlear response. The research requires the design and fabrication of MEMS-resonant gate transistor (RGT) device applied towards the development of MEMS-RGT cochlear biomodel. An array of MEMS-RGT devices can mimic the cochlea by filtering the sound input signals into multiple electrical outputs. The MEMS-RGT device consists of two main components; a) the MEMS bridge gate structure that transduces the sound input into mechanical vibrations and b) the channel with source/drain regions underneath the bridge gate structure that transduce the mechanical vibrations into electrical signals. The CMOS ciruitry both generates signals for rapid adaptation of each sensor, and generates a spike based output suitable for interpretation.
The aim of the project is to develop a new type of MEMS (micro-electro-mechanical systems) microphone with integrated neuromorphic electronics which will directly sense acoustic signals using a number of different MEMS sensors transducing the signal using resonant gate transducers, and conditioning the signals so detected using on-chip electronics.
The idea is (i) to enable the creation of an active microphone that can cope with a wide dynamic range differentially over the spectrum, and (ii) to enable further analyses of the signal using neurobiologically plausible techniques directly.
The key findings are:
1. During the development of MEMS-RGT arrays, the bridge gates have been designed to cover the audible frequency range signals of 20Hz – 20kHz. Aluminium and tantalum have been studied as the material for the bridge gate structure. The downstream etch release technique employing oxygen/nitrogen plasma has been developed to release the bridge gate structure from a sacrificial layer. In the first iteration, aluminium bridge gates have been fabricated. The presence of tensile stress within aluminium had caused the aluminium bridge gates of length >1mm to collapse. In order to address this issue, tantalum bridge gates have been fabricated in the second iteration. Straight tantalum bridge gates in tensile stress and buckled tantalum bridge gates in compressive stress have been characterised. The frequency range of 550Hz – 29.4kHz has been achieved from the fabricated tantalum bridge gates of length 0.57mm – 5.8mm.
2. The channel and source/drain regions have been fabricated and integrated with the aluminium or tantalum bridge gate structures to create the MEMS-RGTs. In this study, the n-channel and p-channel resonant gate transistor (n-RGT and p-RGT) have been considered. The p-RGTs have been found to possess considerably less substhreshold currents than n-RGTs. The threshold voltage, transconductance and substhreshold current for both n-channel and p-channel resonant gate transistor devices have been characterised, where the channel conductance of the n-RGT and p-RGT devices has been modulated successfully and the tuning capability within the audible frequency range has been achieved from the tantalum bridge gates of the p-RGT devices. The characterisation and optimisation of the resonant gate transistor provide the first step towards the development of the adaptive RGT cochlear biomodel for the neuromorphic auditory system application.
3. A spike event coded MEMS-RGT microphone model for neuromorphic auditory systems has been developed. Our microphone system directly converts acoustic signal into bandpassed filtered outputs and encode them as asynchronous spike time events. The microphone system alters its dynamic response by receiving inputs in the spike domain which are then decoded to vary the gate voltage of the MEMS-RGT. The MEMS-RGT sensor model has been simulated and the measurement results from the spike encoder chip for a simulated MEMS-RGT response have been achieved. A set of 10 MEMS-RGT sensors using an etch release process capable of releasing long resonant gate transistor bridges from the sacrificial layer have been fabricated (see point 2).
4. An analogue low-noise MEMS interface circuit which has very small parasitic capacitance at the input node has been designed and fabricated. The circuit is suitable for the MEMS cochlea-mimicking acoustic sensors which are highly parasitic-sensitive due to their low intrinsic sensing capacitance. In order to reduce the electronic noise of the interface circuit, chopper stabilization technique is implemented, and an effective method to optimize the critical transistor size for best noise performance is derived. Simulation results show that, for a MEMS sensing structure with 200 fF static capacitance, the interface circuit achieves a 0.72 aF equivalent capacitance noise floor over 100 Hz to 20 kHz audio bandwidth.
5. An analogue cochlea-mimicking filter has been modelled and is now in the process of being designed in analogue VLSI. The circuit design uses floating active inductors in high-Q and steep cut-off elliptic filters to achieve a response comparable with the biological exemplar. Our implementation offers the advantages of tuning with one variable parameter, high-Q and a steep cut-off response, providing an overall performance that closely matches the biological exemplar.
6. A simulated form of the output of the microphone (which codes the signals from multiple sensors) has been used in an experimnt to test the effectiveness of the data so coded in differentiating different types of musical instruments. This experiment was carried out to assess the effeciveness of the proposed coding technique when applied to real sounds.