Quantcast
Channel: Media & Entertainment Technology » image sensor
Viewing all articles
Browse latest Browse all 3

Low Power CMOS Image Sensor for Mobile and Wearable

0
0

February 23, 2015, ISSCC, San Francisco—Jaehyuk Choi from Samsung Advanced Institute of Technology described a low-power, 15 fps image sensor for always-on mobile and wearable devices. The ability to run on minimal power opens many new applications for smart sensors beyond taking pictures on a mobile device.

This image sensor has two operating modes; ultra-low power always-on low-resolution, and photographic quality with high SNR and resolution. This approach required the 4-T pixels and column circuits be shared, with lightweight processing for always-on functions like face detection at 1 m. the application processor will have to address the post processing requirements.

Power consumption can be easily reduced with lower resolution, from 50+ mW at HD to ~2 mW at HVGA. Reducing bit-depth from 12- to 8-bits decreases power from 9 mW to 1.3 mW. The challenge for further reductions is hindered by the need from the analog circuits for a operating supply voltage greater than 2 V.

The implementation added dynamic voltage scaling and reconfigured the single slope ADC for the photo mode with a SAR ADC. The ADC changes from a high-frequency 12-bit single slope at 3.3 V Vdd to an 8-bit 8-clock cycle SAR ADC at 0.9 V

Another new function is to use charge sharing for the always-on mode so the detector can operate at 0.9 V. The low-Vt transistors in the circuit allow a 0.5 V signal swing and let the column circuits also work at the 0.9 V supply. The converter front end changes from a programmable gain amplifier with a folded-cascode amplifier and comparator to just a comparator.

The circuit performs column and row averaging in always-on mode with digital averaging in non-shared pixels to halve the power requirements, and charge summing with shared pixels. Although the total power was reduced by 50X, the random noise increased when the imager is operating at 0.9 V and the signal swing is reduced. Further voltage scaling below 0.9 V is not possible without changes in the latch circuit and other logic design blocks. The analog circuits can operate down to < 0.8 V with increased SNR.


This design has power consumption of 2.28 mW with a 3.3 V analog supply and 1.8 V digital supply for the photo-grade operation. The sensitivity is 1.1 V/lx-s and random (dark) noise is 0.18 DN. In low-power mode, the imager draws 45.5 uW with 0.9 V for both analog and digital supplies. Sensitivity is 0.3 v/lx-s with a random noise of 0.81DN

An eye tracking app does feature extraction for face detection and face alignment to locate the eyes. In various tests, the app achieved 100 percent eye detection in always-on mode at 0.9 V.
 

 


Viewing all articles
Browse latest Browse all 3

Latest Images

Trending Articles





Latest Images