STT-MRAM characterization and its test implications

Abstract

Spin torque transfer (STT)-magnetoresistive random-access memory (MRAM) has come a long way in research to meet the speed and power consumption requirements for future memory applications. The state-of-the-art STT-MRAM bit-cells employ magnetic tunnel junction (MTJ) with perpendicular magnetic anisotropy (PMA). The process repeatabil- ity and yield stability for wafer fabrication are some of the critical issues encountered in STT-MRAM mass production. Some of the yield improvement techniques to combat the e ect of process variations have been previously explored. However, little research has been done on defect oriented testing of STT-MRAM arrays. In this thesis, the author investi- gates the parameter deviation and non-idealities encountered during the development of a novel MTJ stack con guration. The characterization result provides motivation for the development of the design for testability (DFT) scheme that can help test and characterize STT-MRAM bit-cells and the CMOS peripheral circuitry e ciently. The primary factors for wafer yield degradation are the device parameter variation and its non-uniformity across the wafer due to the fabrication process non-idealities. There- fore, e ective in-process testing strategies for exploring and verifying the impact of the parameter variation on the wafer yield will be needed to achieve fabrication process opti- mization. While yield depends on the CMOS process variability, quality of the deposited MTJ lm, and other process non-idealities, test platform can enable parametric optimiza- tion and veri cation using the CMOS-based DFT circuits. In this work, we develop a DFT algorithm and implement a DFT circuit for parametric testing and prequali cation of the critical circuits in the CMOS wafer. The DFT circuit successfully replicates the electrical characteristics of MTJ devices and captures their spatial variation across the wafer with an error of less than 4%. We estimate the yield of the read sensing path by implement- ing the DFT circuit, which can replicate the resistance-area product variation up to 50% from its nominal value. The yield data from the read sensing path at di erent wafer loca- tions are analyzed, and a usable wafer radius has been estimated. Our DFT scheme can provide quantitative feedback based on in-die measurement, enabling fabrication process optimization through iterative estimation and veri cation of the calibrated parameters. Another concern that prevents mass production of STT-MRAM arrays is the defect formation in MTJ devices due to aging. Identifying manufacturing defects in the magnetic tunnel junction (MTJ) device is crucial for the yield and reliability of spin-torque-transfer (STT) magnetic random-access memory (MRAM) arrays. Several of the MTJ defects result in parametric deviations of the device that deteriorate over time. We extend our work on the DFT scheme by monitoring the electrical parameter deviations occurring due to the defect formation over time. A programmable DFT scheme was implemented for a sub-arrayin 65 nm CMOS technology to evaluate the feasibility of the test scheme. The scheme utilizes the read sense path to compare the bit-cell electrical parameters against known DFT cells characteristics. Built-in-self-test (BIST) methodology is utilized to trigger the onset of the fault once the device parameter crosses a threshold value. We demonstrate the operation and evaluate the accuracy of detection with the proposed scheme. The DFT scheme can be exploited for monitoring aging defects, modeling their behavior and optimization of the fabrication process. DFT scheme could potentially nd numerous applications for parametric characteriza- tion and fault monitoring of STT-MRAM bit-cell arrays during mass production. Some of the applications include a) Fabrication process feedback to improve wafer turnaround time, b) STT-MRAM bit-cell health monitoring, c) Decoupled characterization of the CMOS pe- ripheral circuitry such as read-sensing path and sense ampli er characterization within the STT-MRAM array. Additionally, the DFT scheme has potential applications for detec- tion of fault formation that could be utilized for deploying redundancy schemes, providing a graceful degradation in MTJ-based bit-cell array due to aging of the device, and also providing feedback to improve the fabrication process and yield learning.