Datasets with tag: ECRAM
| Dataset | Description | Authors | Tags | Published | Updated | Date |
|---|---|---|---|---|---|---|
| Parallel training of electrochemical random-access memory | This dataset includes the raw data collected as part of the work in Frontiers of Neuroscience. The ramping_data.csv dataset includes all of the ramping data collected in Fig. 1C,D consisting of 0.25 ms pulses onto all nine ECRAM cells. This dataset plots the raw conductance of the channels in milli-Siemens. To convert to weights, we subtract the values of the offset bias resistor, then divide by 0.05 mS. The bias resistor condutances are 2.16, 2.61, 2.61, 2.26, 2.37, 2.31, 2.55, 2.49, 2.55, each in milli-Siemens. The continuous_seed and discrete_seed dataset contains the training data for the the work. Both proprietary xlsx and open csv formats were used. Column 1 is the training epoch Columns 2-4 are the value of X used for each training step Columns 5-7 are the expected value of Y Columns 8-10 are the computed Z of the training step Columns 11-13 are the length of the digital pulses. For continuous value weight updates, this length is proportional to the error. For discrete value update, this length is either 4 ms (for update) or 0 ms (for no update). Columns 14-20 are the weight values for each ECRAM cells W1-W9. We note that the ECRAM weights are measured AFTER the weight updates have been conducted. These values were plotted in Fig 3A, 3C for seed 1. | Yiyang Li, T Patrick Xiao, Christopher H Bennett, Erik Isele, Hanbo Tao, Matthew J Marinella, Elliot J Fuller, A Alec Talin, Armantas Melianas, Alberto Salleo | ECRAM artifical neural networks hardware accelerator nonvolatile memory neuromorphic computing in-memory computing | 4 years ago | 4 years ago | 2021-04-12 13:31:55 |