Pasar al contenido principal

Resultados relacionados

Resultados totales (Incluyendo duplicados): 2
Encontrada(s) 1 página(s)

Minerva. Repositorio Institucional de la Universidad de Santiago de Compostela
Publicación Capítulo o parte de un libro (BookPart)

Live Demonstration: 5-bit signed SRAM-based DNN CIM for Image Recognition

  • Pereira Rial, Óscar
  • García Lesta, Daniel
  • Vaquero Otal, Lorenzo
  • López Martínez, Paula
  • Brea Sánchez, Víctor Manuel
  • Cabello Ferrer, Diego
This live demonstration shows a mixed-signal Computer In Memory (CIM) macro deep neural network (DNN) integrated circuit in 180 nm CMOS technology for image recognition. Images are coded as pulse width modulation (PWM) signals. DNN weights are stored as voltages in 6T-SRAM memories which drive current sources inside every multiplier. Multipliers are arranged within processing elements laid down in a 2D mesh suitable for image processing. The power consumption per multiplier of the CIM macro is of 0.22 µW, below state-of-the-art competitors following the same multiply and accumulate (MAC) principle.
Minerva. Repositorio Institucional de la Universidad de Santiago de Compostela
Publicación Capítulo o parte de un libro (BookPart)

5-Bit Signed SRAM-Based In-Memory Computing Cell

  • Karimpour, Faranak
  • Pardo, F.
  • García Lesta, Daniel
Hardware accelerators are critical in providing real-time processing for edge computing applications, particu-larly in the context of convolutional neural networks. A crucial challenge in this context is achieving low power consumption while maintaining an appropriate performance in terms of accuracy. This work delves into a thorough analysis of prospective architectures for the core cell of the multiply -and -accumulate function, monitoring each structure's crucial benefits and drawbacks. It includes electrical simulations comparing their performance in a 180 nm process node for 1.8 V and 3.3 V. Moreover, a process corner simulation is proposed to identify on-chip process variations in the voltage error of the proposed design under different input voltages. Notably, the minimum corner errors observed at +15 and -7 sign bits are 0.45 % and 0.63 %, respectively. The significant outcome highlights that the single-switch implementation achieves optimal performance, displaying the lowest error value of 0.14%, specifically at the + 15 sign bit and operating at 1.8 V.