From your very article it is known that the brain is just a physical object.
Any physical object can be simulated.
In the case of the brain we are talking about a quite crude level of abstraction. At the least, quantum and subatomic effects can be ignored. Likewise, gravity, sound, acceleration of the skull, radiation don’t have to be modeled.
It’s likely possible to ignore temperature changes.
And this is the worst case scenario we are talking about. Behavior that’s not identical but not “worse” in any way can surely be produced with much better optimization, such as modelling dissipation as normal distribution and not individual particles, at least once there’s data on the precisely simulated part.

The size of the brain is also quite manageable, especially if we consider that the brain is highly parallel but has a very slow “clock rate” compared to even current processors, and the replacement needn’t fit in the skull, just be a low latency connection.
Finally, the question of whether the future behavior will exactly mirror the replaced part is irrelevant, since the replaced part will be gone, and it’s just the brain/machine interface that has to maintain a consistent/acceptable behavior. They won’t coexist or have to stay in sync.

The missing technology is the neurochemical interface and the gradual replacement, not “can this be done computationally with silicone”.