AxMAP

AxMAP serves as a framework for illustrating the benefits of fast and input-aware error calculation in the efficiency of automatic circuit generation tools. The tool takes five input parameters. Power, area, and delay are considered as the designer's circuit budget constraints. The tolerable Mean Error Distance (MED) over a given pair of input patterns is regarded as the error constraint. AxMAP explores the design space and generates approximate adders using the random search algorithm. In each iteration, the selected design that satisfies the constraints is considered as a valid output. For more details, read our paper on AxMAP [paper].

Setting up and using AxMAP

Requirements

This code has been tested in MATLAB R2015b on a machine with a Corei5 2.5GHz processor and 8 GB of RAM with Windows 10 64-bit serving as the OS. However, using any versions of MATLAB higher than R2015b should be fine.

Steps

1. Clone AxMAP:

git clone https://github.com/mohrez86/AxMAP

2. Open MATLAB, and change the current folder of MATLAB to the AxMAP repository, cloned on your machine.

3. In the MATLAB command window, type edit AxMAP to open the AxMAP.m in the MATLAB editor.

4. In the AxMAP.m file, go to line 14, and set the variable LocalAddress to the directory where you want to save the final Verilog files. For example:

LocalAddress = 'C:\VerilogDirectory\';

5. Save the changes made in step 4, and close the MATLAB editor.

6. In the MATLAB command window, type AxMAP, and press Enter. By running this script, the Verilog description of four adders are produced and stored in the directory saved in the variable LocalAddress (e.g., C:\VerilogDirectory\).

Additional Information

AxMAP comprises several modules which are orchestrated by the AxMAP.m script. There are several variables that are initialized in line 6 to 14 of AxMAP.m:

NumOfDesiredDesigns = 4;
StepIteration       = 1000000;
BitWidth            = 8;
MED_Const           = 50;
POWER_Const         = 50;
AREA_Const          = 10;
InputPatternA       = ones(1, 8)/2;
InputPatternB       = ones(1, 8)/2;
LocalAddress        = 'F:\';

These variables are the parameters of AxMAP, which are explained as follows:

• NumOfDesiredDesigns indicates the number of adders AxMAP generates.
• BitWidth sets the bit-width of the adder circuits.
• MED_Const is a constraint that dictates the maximum amount of error allowed for a generated design, in terms of MED.
• POWER_Const is a constraint that dictates the maximum amount of dynamic power consumption allowed for a generated design.
• AREA_Const is a constraint that dictates the maximum amount of occupied silicon footprint allowed for a generated design.
• InputPatternA and InputPatternB are the input pattern of the first and the second inputs to each design.

Citing AxMAP

You can cite AxMAP by simply copying the follwoing text:

@ARTICLE{8967161,
  author={Rezaalipour, Morteza and Rezaalipour, Mohammad and Dehyadegari, Masoud and Bojnordi, Mahdi Nazm},
  journal={IEEE Transactions on Computers}, 
  title={AxMAP: Making Approximate Adders Aware of Input Patterns}, 
  year={2020},
  volume={69},
  number={6},
  pages={868-882},
  keywords={Adders;Space exploration;Measurement;Monte Carlo methods;Error probability;Computational modeling;Image processing;Approximate computing;adders;mean error distance;error probability;circuit synthesis},
  doi={10.1109/TC.2020.2968905}}

Morteza Rezaalipour, Mohammad Rezaalipour, Masoud Dehyadegari, Mahdi Nazm Bojnordi, AxMAP: Making Approximate Adders Aware of Input Patterns Transactions of Computers, 2020.