This thesis investigates different memory types for use as a synapticstorage in a neuromorphic application for on-chip learning. Our mainconcern was to find a suitable implementation four this purpose. Wewere looking for a memory element which could be used as a distributedstorage with no external control signals or backup. This memory shouldpreferably be analog, which excludes the common digital storagetechniques such as latches and flip-flops. Dynamic multi-level oranalog memory will also be insufficient, since it requires an externalstorage with AD/DA converters to preserve its multi-level or analogvalue. Furthermore, the value stored should be easilyaltered. Previous work has used floating gate(FG), which has manyadvantages in a neuromorphic design, i.e. permanent storage, slowlearning and infinite resolution. However, there are severe deviceproperty mismatches and specialized initialization and programmingtechniques are required to alter the value stored. After initialinvestigation and searches for relevant implementation, no optimalsolutions where found, and we decided to test a novel memory elementwhich is presented in this thesis. It is a multi-level memory, whichstores an analog value over a short period of time. The memorypreserves its own state through a local feedback path, and does notrequire external control signals. The value is easily altered bydirectly applying a voltage, or as done in this thesis, by injectingor drawing a current.
We start with an introduction to neuromorphic electronics anddifferent analog an multi-level memory. We then present the objectiveof the thesis and basic theory. Next, a presentation of the differentcircuit componenets with test results are given. Last, theimplemenetation is discussed and future work is proposed.