A 0.5 V 55 $\mu \text{W}$ 64 $\times $ 2 Channel Binaural Silicon Cochlea for Event-Driven Stereo-Audio Sensing
Citations
275 citations
Cites background from "A 0.5 V 55 $\mu \text{W}$ 64 $\time..."
...Recent work at INI includes the development of neuromorphic vision sensors [8], silicon cochlea [9], and mediumscale neuromorphic processors such as the Reconfigurable On-Line Learning Spiking (ROLLS) and cxQuad chips....
[...]
99 citations
90 citations
Cites background from "A 0.5 V 55 $\mu \text{W}$ 64 $\time..."
...To mimic biology more closely, neuromorphic visual [178], [179], auditory [180] and tactile [181] sensors have been developed which operate in an asynchronous, spike based manner similar to human retina, cochlea and mechanoreceptors, respectively....
[...]
87 citations
87 citations
Cites background or methods from "A 0.5 V 55 $\mu \text{W}$ 64 $\time..."
..., 2015) and the Dynamic Audio Sensor (DAS) (Liu et al., 2014; Yang et al., 2016) fall roughly into two categories: either by the use of neural network methods or machine learning algorithms....
[...]
...INTRODUCTION The event processing methods for the asynchronous spikes of event-based sensors such as the Dynamic Vision Sensor (DVS) (Lichtsteiner et al., 2008; Berner et al., 2013; Posch et al., 2014; Yang et al., 2015) and the Dynamic Audio Sensor (DAS) (Liu et al., 2014; Yang et al., 2016) fall roughly into two categories: either by the use of neural network methods or machine learning algorithms....
[...]
...…Dynamic Vision Sensor (DVS) (Lichtsteiner et al., 2008; Berner et al., 2013; Posch et al., 2014; Yang et al., 2015) and the Dynamic Audio Sensor (DAS) (Liu et al., 2014; Yang et al., 2016) fall roughly into two categories: either by the use of neural network methods or machine learning algorithms....
[...]
...Time-binned SC features have been used for the speaker identification task using spike recordings generated from the TIMIT dataset (Liu et al., 2010; Li et al., 2012), the YOHO dataset (Chakrabartty and Liu, 2010), and real-world DAS recordings (Anumula et al., 2017)....
[...]
...The methods evaluated in this work were carried out on recordings from the CochleaAMS1b and CochleaAMS1c, while they will be evaluated on CochLP in the future....
[...]
References
3,253 citations
"A 0.5 V 55 $\mu \text{W}$ 64 $\time..." refers background in this paper
...The benefits of this type of clockless system are threefold: 1) feature extraction in analog domain can save power, particularly when the required feature SNR is low to medium [16], [17]; 2) asynchronous event encoding helps reduce data redundancy, because the output event rate is proportionally dependent on the input activity and becomes zero if the input is quiescent [18], [19]; and 3) event-driven processors, such as continuous-time (CT) DSPs [19] and spiking neural networks [20], have adaptive power consumption that is correlated with the incoming event rate, and their processing latency can be reduced because of low data redundancy; low processing latency is particularly important for real-time applications involving human–machine interactions, where the machine’s immediate reaction to human expressions, such as speech, is often necessary....
[...]
2,460 citations
"A 0.5 V 55 $\mu \text{W}$ 64 $\time..." refers background in this paper
...6) with maximum 18 dB attenuation is used to extend the system upper bound input signal range while keeping the distortion of the following BPF low; inclusion of this attenuator is further motivated by the fact that active microphones can produce large amplitude output voltages with close-by loud sound even without standalone preamplifiers, which may cause large-signal oscillation of BPFs at high Q [43]....
[...]
767 citations
695 citations
"A 0.5 V 55 $\mu \text{W}$ 64 $\time..." refers background in this paper
...in the near-threshold region [56], [57] or completely avoided in an embedded system where the event streams are directly sent in parallel to the back-end event-driven processor on the same chip, as illustrated in Fig....
[...]
495 citations
"A 0.5 V 55 $\mu \text{W}$ 64 $\time..." refers background in this paper
...The benefits of this type of clockless system are threefold: 1) feature extraction in analog domain can save power, particularly when the required feature SNR is low to medium [16], [17]; 2) asynchronous event encoding helps reduce data redundancy, because the output event rate is proportionally dependent on the input activity and becomes zero if the input is quiescent [18], [19]; and 3) event-driven processors, such as continuous-time (CT) DSPs [19] and spiking neural networks [20], have adaptive power consumption that is correlated with the incoming event rate, and their processing latency can be reduced because of low data redundancy; low processing latency is particularly important for real-time applications involving human–machine interactions, where the machine’s immediate reaction to human expressions, such as speech, is often necessary....
[...]