The Power Knee prosthesis, designed for transfemoral amputees, is based on advanced artificial intelligence and nifty mechanics. Ossur, the Icelandic manufacturer of this device, explains:

The Power Knee combines an electromechanical power source with a ‘Sound-Side Sensory-Control’ (SSSC), taking another step towards the mimicking of kinetic and kinematic trajectories and the replication of physiological processes.
The SSSC is a unique way to restore proprioception. The human body requires this indispensable physiological process in order to cope with the broad range of human locomotory functions. This technology exceeds previously established knee systems by gathering sensory information one step ahead of the prosthesis. Whereas other prosthetic solutions fulfill the role of a reactive device, the Power Knee is proactive.
Guided by an advanced type of artificial intelligence the electromechanical power source has the ability to replace muscle function around the knee-joint. Whereas previously established knee systems have been limited to the imitation of excentric muscle work, the Power Knee also replaces concentric muscle activity. The result is outstanding function, unmatched by any other product, allowing the user to ascend stairs and cover longer distances, foe example, with significantly reduced effort.
The Power Knee system monitors an individual’s gait pattern by sampling force and angular measurements at a rate of 1350 times per second. This level of precision is required to achieve an accurate user-system symbiosis.
The combination of the unequalled torque and power generation, the Sound-Side Sensory-Control and the advanced artificial intelligence, rightfully position the Power Knee as a pioneering product on Ossur’s Bionic platform.

More info at Ossur… (don’t forget to watch those videos!)
(hat tip: Gizmodo)

The MIT Technology Review describes the research behind the first direct electrical interface between a semiconductor device and an individual mammalian nerve cell:

Context: The neurons of the mammal brain are hard to study, even when they’re isolated in the lab. For more than a decade, scientists have analyzed the large neurons of leeches and snails by linking them directly to silicon chips that record their electrical activity. But mammalian neurons are smaller, and though they can be grown on silicon, the resulting signals are typically too weak to yield useful data. The electrical activity of mammalian brain cells can be read with electrodes, but that can be imprecise and requires careful preparation steps.
Moritz Voelker and Peter Fromherz at the Max Planck Institute for Biochemistry have now designed the first computer chip that can record the firing of mammalian neurons, though so far only in a petri dish.
Methods and Results: As a neuron fires, the voltage across it changes, so a neuron on a chip affects how transistors underneath it conduct electricity. But in chips with conventional transistor designs, there’s so much naturally occurring noise that it swamps neural signals. So Voelker and Fromherz changed the geometry of the transistors to suit the electrical properties of living neurons. They buried the conducting channels of their transistors a few nanometers deeper than usual, making the transistor more sensitive to the low voltages and firing speeds of neurons. The transistors could detect the signal of an ?individual rat neuron in a group, without the elaborate sample preparation that ?conventional electrodes require. What’s more, the tran?sistors are significantly smaller than individual neurons and could in principle provide information on how subsections of a neuron behave.

The abstract at the Max Planck Institute for Biochemistry …
Flash: BrainGate Neural Interface System…