On the impact of 3D integration on high-throughput sensor information processing: A case study with image sensing

This paper presents a study for integrating a wavelet based image compression system in a 3D stack for wireless sensor node application. The high-bandwidth die-to-die interface from the vertical stacking of the sensor, the image buffer, and the signal-processing unit (SPU) is exploited for parallel processing. An image is divided into multiple segments, where each segment is compressed independently. However this method can negatively impact the compression ratio due to the length extension at the border of each segment. This paper explores the design space of the multi-segment architecture for the wavelet based image compression, including the number of segments, the operating frequency, and the supply voltage to reduce the power dissipation, and maintain a target image rate under varying channel condition. The effectiveness of the multi-segment implementation with 2D- versus 3D- integration is compared.