A 100¿W 64×128-Pixel Contrast-Based Asynchronous Binary Vision Sensor for Wireless Sensor Networks

Abstract

Raster-scan architectures are mostly oriented toward quality image reproduction [1], but they do not meet some basic requirements,that are crucial for battery-operated sensors for sensor network applications, such as low-energy visual features extraction and control strategies for output bandwidth reduction [2]. We present a 64×128-pixel vision sensor, whose pixels estimate and perform a 1b quantization on the local contrast with a low energy budget. The pixel-embedded time-adaptive visual processing is based on a charge-transfer mechanism, featuring no DC power consumption [3]. The asynchronous readout process takes 147µs and dispatches the column address of each asserted pixel, significantly reducing the chip activity at the interface. For typical indoor visual contrast estimation, involving 5% of the total number of pixels, the sensor exhibits a power consumption of 100µW at 3.3V and 50 frames/s. Fig. 32.7.1. The imager is organized into an addressable array of pixels, with a 64-cells, a ROW DECODER and a COLUMN DECODER made of 128 FIND cells. The raster-scan readout process is completely asynchronous and only dispatches the addresses of the pixels detecting a contrast. This drastically reduces both data bandwidth and power consumption. After scanning the imager, the sensor stops, ready for a new acquisition process. Contrast estimation [4] is implemented using a kernel of three pixels (O, N, E). After reset, the pixels exhibit a voltage discharge ramp, due to the impinging light. The contrast estimation process (Fig. 32.7.2) starts when the most illuminated pixel (VPixN) reaches the threshold V TH1 (t 1 ) and stops at t 2 , when the same threshold is crossed by the least illuminated pixel (VPixO). Within t 1 and t 2 , the voltage drop between VPixO and VPixN is tracked and the final value (V EDGEO ) is sampled and quantized to 1b: V EDGEO = VPixO(t2) -VPixN(t2). A block diagram of the chip is shown inThe n-well/p-sub pixel (Fig. 32.7.3) drives a starved comparator (Comp1) and one of the three inputs of the contrast block (CB) [4]. The CB is also fed by the other two analog signals (VPixN, VPixE) of the neighboring photodiodes (PixelN, PixelE) and by the three comparators (OutO, OutN, OutE). The most illuminated photodiode (VPixN), with the lowest voltage, drives node A, through the source-follower Mp2, discharging Mc1 to V SS (OutO = OutN = OutE = H). VPixN reaches V TH1 at t 1 , breaking the path to V SS (OutN = L), thus allowing the charge to be accumulated on Mc2. The charge transfer process continues until t 2 , the time at which VPixO crosses the threshold. Here, the charge path from Mc1 to Mc2 (OutO = OutN = OutE = L), is turned off and the contrast voltage is held on Mc2 with a gain of about 3, depending on the ratio between the two capacitors:The contrast is estimated dynamically, with no DC power consumption except leakage and photo-generated currents. The contrast block requires 770pJ/pixel, corresponding to 33nW/pixel at 50fps. Comp2 compares V EDGEO with V TH2 and its binary output is either store...

Activity-driven, event-based vision sensors

“…[38][39][40][41]) and high speed robotics [42,43]. Eight papers on event based sensing have been accepted to the very competitive IEEE International Solid State Circuits Conference since 2003 (two on auditory sensors [44,45] and six on vision sensors [9,16,27,33,46,47]), showing that this approach is starting to impact mainstream electronics. We expect that the next few years will bring substantial progress in this approach to vision.…”

“…It has three operating modes-TD, SD and intensity readout-all at a low power consumption of 1mW, making it attractive for wireless sensor networks, like [27].…”

“…Year 2001 2003 2005 2006 2008 2009 2010 2010 2010 Source Zaghloul, Boahen [30] Rüedi et al [16] Mallik et al [9,32] Lichtsteiner et al [5,33] Massari et al [27] Ruedi et al [31] Posch et al 2010 [34] LinaresBarranco et al [2] Culurciello et al [3] Functionality Asynchronous spatial and temporal contrast, …”

STDP and STDP variations with memristors for spiking neuromorphic learning systems

“…For example, a large collection of spike-driven vision sensors have been reported, such as sensors for luminance (Culurciello et al, 2003; Chen et al, 2011), temporal contrast (Barbaro et al, 2002; Mallik et al, 2005; Chan et al, 2007a; Lichtsteiner et al, 2008; Leñero-Bardallo et al, 2011; Posch et al, 2011; Serrano-Gotarredona and Linares-Barranco, 2013), motion (Kramer, 1996; Sarpeshkar et al, 1996; Ozalevli and Higgins, 2005), and spatial contrast (Ruedi et al, 2003; Zaghloul and Boahen, 2004; Costas-Santos et al, 2007; Massari et al, 2008; Leñero-Bardallo et al, 2010). Spike-driven principles have also been used for auditory systems (Sarpeshkar et al, 2005; Wen and Boahen, 2006, 2009; Chan et al, 2007b), competition and Winner-Take-All networks (Indiveri, 2000; Chicca et al, 2007; Oster et al, 2008), learning (Mill et al, 2011), classification (Mitra et al, 2009), fall detection (Fu et al, 2008), and systems distributed over wireless sensor networks (Teixeira et al, 2005; Massari et al, 2008). Apart from real-time sensing, spike-driven processing systems can produce extremely fast responses.…”

Area and power reduction techniques for time-based image sensor pixel design

“…Additionally, pixel-level processing allows the implementation of various preprocessing tasks like frame differencing [4] or edge detection [5] in a time and power efficient way.…”