You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Following on from a discussion with @nandhinic, We need the ability to inject into SRAMs. There are two working examples of how this should work:
A shadow SRAM that stores a bit flip mask. On a read, you read one word from the normal SRAM and one word from the shadow SRAM and xor them together.
(1) augmented with a probability. This adds one configurable register that stores a bit flip probability. On an SRAM read, this determines the probability of a bit flip. Note: this overlays a uniform probability across the entire SRAM.
At this point, pre-bit IID probabilities are possibly desirable, but very expensive. This, like with the existing LFSR injector requires one word storing a probability for each bit flip. If the probabilities are stored as 32-bit integers, then this would mean a 32x increase in area.
The text was updated successfully, but these errors were encountered:
Following on from a discussion with @nandhinic, We need the ability to inject into SRAMs. There are two working examples of how this should work:
At this point, pre-bit IID probabilities are possibly desirable, but very expensive. This, like with the existing LFSR injector requires one word storing a probability for each bit flip. If the probabilities are stored as 32-bit integers, then this would mean a 32x increase in area.
The text was updated successfully, but these errors were encountered: