Research And Design Of Computing-in-memory Circuit Based On Variable Gain Amplifier

The well-known Moore's Law is about to end after CMOS devices using 7nm process technology.It is becoming more and more difficult to improve the performance of the chip and reduce the cost of the chip by reducing the feature size of the device.At the same time,with the development and rise of technologies such as artificial intelligence,the Internet of Things,and big data,data has grown rapidly in the form of an almost exponential explosion.The computer system of the von Neumann architecture,which has dominated the computing field for decades,has problems of memory walls and power walls,which is called the Von Neumann bottleneck.The storage wall problem refers to the mismatch between the speed of the computing unit and the storage unit.The speed of the former is much faster than the latter.The speed of the storage unit has become a bottleneck that affects the speed of calculation.The power wall problem refers to the frequent transmission of data between physically separated computing units and storage units that consumes a lot of power.In order to solve the Von Neumann bottleneck problem,the internal calculation circuit,also known as the integrated circuit of storage and calculation,has been proposed.Computing-In-memory circuit,as the name suggests,refers to a circuit that directly uses memory for data calculation by fusing data storage and calculation together,avoiding the storage wall problem and power consumption wall problem caused by the separation of the computing unit and the storage unit.It solves the Von Neumann bottleneck of von Neumann architecture computers.Currently,computing-in-memory circuits are mainly based on two types of memory,non-volatile memory and volatile memory.The computing-in-memory circuit proposed in this paper is based on a new type of nonvolatile memory-flash memory(Flash)and uses a variable gain amplifier as the calculation result of the analog signal processing circuit.Non-volatile memory can provide inherent computing power,can achieve data storage and calculation in the same physical unit.In addition,since the non-volatile memory can save data after power off,it can be directly used in the system on chip without the need for additional off-chip memory.In deep convolutional neural network(CNN),more than 90% of the computation is convolution computation,which can be realized quickly and with low power consumption by using computing in memory circuit.Main innovation point of this article is based on the 1024×512 size FLASH array implementation of bit 4 signal operation parallelly,and put forward a new kind of based on FLASH memory computing architecture,which is contained in the FLASH array,and transimpedance amplifiers,variable gain amplifier and SAR-ADC.Because of the use of variable gain amplifier,the memory calculation circuit can meet the needs of different multiplication and addition combination and convolution operation speed.