IMAC-Sim is a Python-based simulation framework, which creates the SPICE netlist of the In-Memory Analog Computing (IMAC) circuit based on various device- and circuit-level hyperparameters selected by the user, and automatically evaluates the accuracy, power consumption and latency of the developed circuit using a user-specified dataset. The list of currently supported inputs are as follows.
- data_dir: The directory where data files are located
- spice_dir: The directory where spice files are located
- dataset_file: Name of the dataset file
- label_file: Name of the label file
- weight_var: percentage variation in the resistance of the synapses
- testnum: Number of input test cases to run
- testnum_per_batch: Number of test cases in a single batch
- firstimage: Starting point of the test inputs in the dataset file
- vdd: The positive supply voltage
- vss: The negative supply voltage
- tsampling: The sampling time in nanosecond
- nodes: Network Topology, an array which defines the DNN model size
- hpar: Array for the horizontal partitioning of all hidden layers
- vpar: Array for the vertical partitioning of all hidden layers
- gain: Array for the differential amplifier gains of all hidden layers
- tech_node: The technology node e.g. 9nm, 45nm etc.
- metal: Width of the metal line for parasitic calculation
- T: Metal thickness
- H: Inter metal layer spacing
- L: length of the bitcell
- W: width of the bitcell
- D: Distance between I+ and I- lines
- eps: Permittivity of oxide
- rho: Resistivity of metal
- rlow: Low resistance level of the memristive device
- rhigh: High resistance level of the memristive device
Follow this paper for more information. Md Hasibul Amin, Mohammed E. Elbtity, and Ramtin Zand. 2023. IMAC-Sim: A Circuit-level Simulator For In-Memory Analog Computing Architectures. In Proceedings of the Great Lakes Symposium on VLSI 2023 (GLSVLSI '23). Association for Computing Machinery, New York, NY, USA, 659–664. https://doi.org/10.1145/3583781.3590264
- First go to the directory data and put the input test data, label data, pre-trained weights and biases into the directory. Sample files are provided for a 400 X 120 X 84 X 10 DNN model.
- Go to the directory spice and put the neuron subcircuit, differential amplifier subcircuit and transistor models into the directory. Sample model files are provided.
- Open testIMAC.py and modify the list of inputs as required.
- Make sure you have HSpice installed in your machine and accessible through hspice command in the terminal. The code initiates a HSpice run through an os.system call of the hspice command.
- Run the IMAC using the command python testIMAC.py. This will build the necessary IMAC subcircuits in the spice directory, run the netlist on HSpice and print the accuracy and power consumption results of your input test cases batch-by-batch in the terminal.
The following papers use IMAC-Sim.
- M. H. Amin, M. E. Elbtity and R. Zand, "Xbar-Partitioning: A Practical Way for Parasitics and Noise Tolerance in Analog IMC Circuits," in IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 12, no. 4, pp. 867-877, Dec. 2022, doi: 10.1109/JETCAS.2022.3222966.
- M. Elbtity, A. Singh, B. Reidy, X. Guo and R. Zand, "An In-Memory Analog Computing Co-Processor for Energy-Efficient CNN Inference on Mobile Devices," 2021 IEEE Computer Society Annual Symposium on VLSI (ISVLSI), Tampa, FL, USA, 2021, pp. 188-193, doi: 10.1109/ISVLSI51109.2021.00043.
- Md Hasibul Amin, Mohammed Elbtity, Mohammadreza Mohammadi, and Ramtin Zand. 2022. MRAM-based Analog Sigmoid Function for In-memory Computing. In Proceedings of the Great Lakes Symposium on VLSI 2022 (GLSVLSI '22). Association for Computing Machinery, New York, NY, USA, 319–323. https://doi.org/10.1145/3526241.3530376
- M. H. Amin, M. Elbtity and R. Zand, "Interconnect Parasitics and Partitioning in Fully-Analog In-Memory Computing Architectures," 2022 IEEE International Symposium on Circuits and Systems (ISCAS), Austin, TX, USA, 2022, pp. 389-393, doi: 10.1109/ISCAS48785.2022.9937884.