Hello, I have a physical optical chip want to train, and I want to use your ONN architecture first to do some training and evaluation like MNIST task. But the problem is that our chip's nullification method is RELLIM rather than PRELLIM used in your model, because we haven't added noninvasive CLIPP detectors at the end of the node output port, it is complex in fabrication. So, to make accurate match between your ONN model with our physical chip, I want to know how can I change the way from PRELLIM to RELLIM, because PRELLIM has robustness to static and dynamic errors which should be showed in the real chip. And the effect of fabrication error is important for final task accuracy when the chip is using RELLIM for nullification.
Thanks.
Hello, I have a physical optical chip want to train, and I want to use your ONN architecture first to do some training and evaluation like MNIST task. But the problem is that our chip's nullification method is RELLIM rather than PRELLIM used in your model, because we haven't added noninvasive CLIPP detectors at the end of the node output port, it is complex in fabrication. So, to make accurate match between your ONN model with our physical chip, I want to know how can I change the way from PRELLIM to RELLIM, because PRELLIM has robustness to static and dynamic errors which should be showed in the real chip. And the effect of fabrication error is important for final task accuracy when the chip is using RELLIM for nullification.
Thanks.