Bio-inspired processing of multi-dimensional information

本课题组利用纯二进制实现了包括时域信息和二维空间信息的多维信息的硬件实现。多值、模拟和模拟数字合并系统。特别关注的是生物启发的系统，如适应性学习系统，其中刺激的反应是由过去的经验和重组系统改变，并期待基于联想的想法或粗略的判断快速反应。在研究课题中，自适应学习神经芯片的制造使用铁电门FET。通过将CMOS施密特触发电路与具有SOI（绝缘体上硅）结构上的SrBi2Ta2O9膜的铁电栅FET集成来制造测试芯片，并且证明输出脉冲频率随着输入脉冲数的增加而增加。这种操作可以看作是一种自适应学习功能。一种是由电容和npn双极晶体管组成的外部时钟型，另一种是由非稳态CMOS多谐振荡器组成的内部时钟型.在模数融合电路方面，实现了智能图像处理器的脉宽调制系统，其中功能图像传感器、模式匹配处理器等集成在芯片中。在CMOS数字和神经MOS技术。研制了高精度模拟非易失性存储器和寻找最相似数据的关联引擎芯片。最后，研制了一种新型的视网膜边缘检测芯片，该芯片除光电池区外的整个区域作为MOSFET的沟道区，起到二维布线的作用。这种布局在避免光电池之间互连的复杂性方面是有用的。

In this research group, hardware implementation of multi-dimensional information which includes time domain information as well as two-dimensional special information is realized using pure-binary. multi-valued, analog and analog-digital-merged systems. Particular attention is paid in such bio-inspired system as the adaptive-learning system in which response for stimulus is changed by the past experience and the reorganizing system, and quick response based on association of ideas or rough judgement is expected.In the research topic in which an adaptive-learning neurochip is fabricated using ferroelectric-gate FETs. a test chip was fabricated by integrating a CMOS Schmitt trigger circuit with ferroelectric-gate FETs with SrBi2Ta2O9 film on an SOI (silicon-on-insulator) structure and it was demonstrated that the output pulse frequency increased with increase of input pulse number. This operation can be regarded as an adaptive-learning function.Concerning generation of caotic signals, two kinds of integrated circuits were fabricated ; one is the external clock typewhich is composed of capacitors and npn bipolar transistors and the other is internal clock type which is composed of astable CMOS multi-vibrators.Concerning analog-digital-merged circuits, pulse width modulation system was imp lemented for intelligent imaging processor, in which functional image sensors, pattern matching processors, and so on are integrated in a chip. In CMOS digital and neuron MOS technologies. high-precision analog nonvolatile memories and association engine chip to look for the most resemble data were developed.Finally, a new retina chip for edge detection was fabricated, in which the whole area except for the photo-cell area acts as the channel region of MOSFET and plays a role of two-dimensional wiring. This layout is useful in avoiding the complexity of interconnection among photo-cells.