The growth of AI services requires increased compute capabilities on edge devices. Neuromorphic techniques are critically important for obtaining the levels of efficiency needed to achieve this transition. BrainChip has been a pioneer in bringing digital, portable, event-based neuromorphic solutions to market. The second generation of Akida, BrainChip’s neural processor, adds additional efficient intelligent processing for multidimensional streaming data with Temporal Event-based Neural Networks (TENNs) and vision transformer acceleration in hardware. This enables much more capable, portable, passively cooled devices that can perform advanced processing for vision, voice, vibration and other modalities close to the sensor, without need for cloud connections, serving a wide variety of applications from autonomous drones to health wearables. In this talk, we discuss how the natural compression offered by TENNs algorithms can radically reduce model size and computation without compromising accuracy for more advanced multimodal applications that use large language models and large vision models.