Approximate Computing exploits the inherent error resilience of many applications to optimize power consumption, run time, and/or chip area. In particular in audio, image and video processing, but also in data mining, scientific computing or resource allocation tasks, approximate results are “good enough” and hard to distinguish from perfect results.In the past few years, approximate computing has been addressed from various directions and received increasing attention. The goal of this session is to provide a multi layer view of the challenges in test and fault tolerance of Approximate Computing systems.

Room: Florentine II
Organizer: Hans-Joachim Wunderlich (University of Stuttgart)
Moderator: Sybille Hellebrand (University of Paderborn)

• Introduction to Approximate Computing
  Jie Han (University of Alberta, Canada)

  Abstract: Approximate computing has emerged as a new paradigm for energy-efficient design of circuits and systems. In this talk, we start with an introduction to approximate computing and focus on the recent progress on approximate circuit design. These designs include arithmetic circuits of adders, multipliers and dividers, among others. A brief classification is presented and the features of different classes of designs are summarized.

• Fault Tolerant Approximate Computing Using Emerging Non-Volatile Spintronic Memories
  Mehdi Tahoori (Karlsruhe Instiute of Technology)

  Abstract: The approximate computing paradigm can be leveraged to exploit the intrinsic error resilience of many applications to enable highly efficient memory implementations. Most emerging memory technologies have a wide range of performance/power/reliability trade-offs which can be exploited in fault tolerant approximate computing. In this work, we use approximate computing to tolerate the increased retention failure rate caused by relaxing the thermal stability factor of Spin-Transfer Torque Magnetic RAM (STT-MRAM) to enable fast and energy efficient STT-MRAM cache memories.

• Fault Tolerance of Approximate Compute Algorithms
  Hans-Joachim Wunderlich (University of Stuttgart)

  Abstract: Algorithms and applications running on approximative hardware may tolerate deviations and errors up to a certain degree. Online test and fault tolerance at this level are not just a binary go/nogo decision, but require more sophisticated error detection mechanisms to classify results still as acceptable. This talk will present application level fault tolerance techniques to be used in scientific computing and statistical methods for distinguishing errors and faults from acceptable results.