Abstract
In the last few years, dynamically configurable approximate multipliers have been explored to tune the energy-quality trade-off in error-tolerant applications at runtime. Typically, the multiplier accuracy is adjusted by adding a constant correction factor equal to the multiplier mean error to the result, which is found offline assuming a predetermined input distribution. This paper describes a simple approach to update the correction term at runtime, thus adapting it to the actual incoming inputs. It takes advantage of the spatial and/or temporal correlation typically shown by input data in error-tolerant applications, such as image and video processing. When applied to a typical case study implemented with a commercial UTBB FDSOI 28 nm technology, the proposed approach shows an energy reduction of up to 34% at iso-quality and a quality improvement of up to +9 dB, −4× and +35% at iso-energy, in terms of peak-to-noise ratio (PSNR), normalized error distance (NED) and structural similarity index metric (SSIM) respectively, compared to the traditional technique based on a constant correction factor.
Highlights
IntroductionApproximate computing consists in relaxing the constraint of an exact computation in order to trade the quality of the result with speed, area and power consumption [1,2]
Configurable ApproximateApproximate computing consists in relaxing the constraint of an exact computation in order to trade the quality of the result with speed, area and power consumption [1,2].As fundamental arithmetic blocks in signal processing, approximate multipliers have been widely explored in the last few years [3–15]
The value of the correction constant is chosen at design time and it is equal to the mean error of the approximate multiplier, assuming a certain statistical distribution of the inputs [6]
Summary
Approximate computing consists in relaxing the constraint of an exact computation in order to trade the quality of the result with speed, area and power consumption [1,2]. Several approximate techniques have been proposed, such as column truncation [5,6], approximate compressors [7,8], the use of error-tolerant adders [9], input truncations [10], vertical and horizontal cut [12] and input encoding [13,14] All these techniques exploit a simple error-correction technique, such as adding an error compensation constant to the approximate result in order to increase the accuracy [15]. The value of the correction constant is chosen at design time and it is equal to the mean error of the approximate multiplier, assuming a certain statistical distribution (typically uniform) of the inputs [6]. This paper investigates the ability to exploit the dynamic configurability of such multipliers in order to dynamically adapt the error compensation constant to the incoming inputs over time This is carried out by periodically switching the multiplier operation mode between two different accuracy levels and updating the correction factor in each period.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
