The present disclosure provides a domain wall magnetic tunnel junction device. Integration of input spikes pushes a domain wall within a ferromagnetic track toward a magnetic tunnel junction (MTJ). An energy gradient within the track pushes the domain wall away from the MTJ by leaking accumulated energy from the input spikes. If the integrated input spikes exceed the energy leak of the gradient within a specified time period, the domain wall reaches the MTJ and reverses its resistance, producing an output spike. The leaking energy gradient can be created by a magnetic field, a trapezoidal shape of the ferromagnetic track, or nonuniform material properties in the ferromagnetic track. A feedback and feedforward crossbar configuration is disclosed that provides for a neuromorphic computing system suitable for unsupervised learning.