Autonomous vehicles are still in their research phase. There are many possibilities for perfection in their functional and non-functional behaviour. A formal analysis, at an early stage, will help to identify various inconsistencies in the design of autonomous vehicles. In this work, various possible risks of an autonomous car are identified. A risk taxonomy is created based on the major components of an autonomous car. Architecture Analysis and Design Language (AADL) has been used for the formal risk modelling of various components of an autonomous car. Further, the fault tree method is used for risk analysis. Identifying software risks of autonomous cars in the early phase is helpful for design optimization and safety enhancement.