We propose a semiparametric functional single index model to study the relationship between a univariate response and multiple functional covariates. The parametric part of the model integrates the functional linear regression model and the sufficient dimension reduction structure. The nonparametric part of the model allows the response-index dependence or the link function to be unspecified. The B-spline method is used to approximate the coefficient function, which leads to a dimension folding type model. A new kernel regression method is developed to handle the dimension folding model, which allows the efficient estimation of both the index vector and the B-spline coefficients. We also establish the asymptotic properties and semiparametric optimality for the estimators.