Symmetric phase-change memory devices

A neural network processing system, comprising:

• a first layer of one or more first neurons;
• a second layer of one or more second neurons;
• an array of weights that accepts a voltage output by the first layer and outputs a current to the second layer, wherein each weight in the array of weights includes:
• a first phase change memory device connected to a bitline, comprising a first material that increases resistance when a set pulse is applied;
• and
• a second phase change memory device connected to the bitline in serial with the first phase change memory device, comprising a second material, different from the first material, that decreases resistance when a set pulse is applied.

A neural network processing system, comprising:

• a first layer of one or more first neurons;
• a second layer of one or more second neurons;
• an array of weights that accepts a voltage output by the first layer and outputs a current to the second layer, wherein each weight in the array of weights includes:
• a first phase change memory device, comprising a first material that increases resistance when a set pulse is applied;
• a second phase change memory device, comprising a second material, different from the first material, that decreases resistance when a set pulse is applied, wherein the first phase change memory device and the second phase change memory device are connected in series to a bitline;
• a first bypass switch configured to short-circuit the first phase change memory device when enabled;
• and
• a second bypass switch configured to short-circuit the second phase change memory device when enabled, wherein control lines to the first bypass switch and the second bypass switch are connected to a wordline.