We consider the effect of noise in sparse Boolean Networks with redundant functions. We show that they always exhibit a non-zero error level, and the dynamics undergoes a phase transition from non-ergodicity to ergodicity, as a function of noise, after which the system is no longer capable of preserving a memory of its initial state. We obtain upper-bounds on the critical value of noise for networks of different sparsity. PACS numbers: 89.75.Da,05.65.+b,91.30.Dk,91.30.Px Introduction-Biological systems are unavoidably noisy in their nature, but often need to function in a predictable fashion [1]. In such a situation, strategies to diminish the harmful effect of noise will significantly impact the fitness of a given organism. The most fundamental protection mechanism a system can adopt is the redundancy of its underlying components, since the resulting coincidences necessary to impact the proper function of the system can drastically diminish the probability of error. In this letter we are concerned with the effect of redundancy in gene regulation; in particular in a simple Boolean Network (BN) model. We assume that each component in the system is arbitrarily redundant, with the only restriction that the number of inputs per component is fixed and finite. In a general manner, we are able to show that redundancy can always guarantee reliable dynamics, up to a given critical value of noise, above which the system is incapable of maintaining any memory of its past states. From simple considerations, we are able to obtain upper bounds on the maximum resilience attainable. This provides an important frame of reference to determine the reliability of a system with a given sparsity.We begin by defining the model, and how noise is introduced. A Boolean Network (BN) [2] is a directed graph with N nodes, representing the genes, which have an associated Boolean state σ i ∈ {0, 1}, corresponding to the transcription state, and a function f i ({σ j } i ), which determines the state of node i given the states of its input nodes {σ j } i . The number of inputs of a given node is k i , or simply k if its the same for all nodes. This system is usually updated in parallel, such that at each time step t, we have σ i (t + 1) = f i ({σ j (t)} i ). Starting from an initial configuration, the system will evolve, and eventually settle on an attractor. In a real system, the expression level of a particular gene can fluctuate, despite the stability of its input states [3]. This characteristic can be incorporated qualitatively in the BN model as uniform noise [4][5][6][7][8][9][10], defined as a probability p that, at each time step, the value of a given input σ j ∈ {σ j } i of a node i is flipped, prior to the evaluation of the function f i . The value of p plays the role of a temperature in the system. If p = 0 the original deterministic model is recovered, and if p = 1/2 the system becomes effectively decoupled, with entirely stochastic dynamics.In the model above, it is known that error resilience does not spontaneously emerge, sinc...