Methods for measuring RRAM basically include DC sweep, PD test, and various other tests. In this article, we will examine the tests that measure the characteristics of the device.
For a basic explanation of RRAM, please refer to the previous article.
Measurement method
In the previous article, I explained that RRAM has a simple MIM structure. Of course, to implement an RRAM device as a circuit, it must be made into a structure called a crossbar array. However, to understand the characteristics of a single device, a device can be made with a simple structure like the above. Simply cut a bare wafer (1 cm x 1 cm) to an appropriate size, place BE (Bottom Electrode) and oxide, and then pattern and deposit TE (Top Electrode) on top. That's it.
In order to measure, you need to make a probe contact, right?? For TE, you can make a direct contact, and for BE, you scratch the oxide that is deposited all over with a diamond cutter to expose the BE, attach an indium ball to make a contact pad, and then make contact with the probe tip to proceed with the measurement.
DC sweep
I may explain this separately later, but for the component measurements, I used a program I created. Our lab uses a tool called Labview, which controls a Keithley 236 SMU (Source Measure Unit) to perform DC sweep measurements.
By setting voltage as source and current as result, you can obtain the I-V curve of the device.
Forming process
To operate the device, it must be formed. Before forming, the device's resistance is extremely high, so little current flows and the Forming Vis also very high. After forming, a reset is performed and then a Set/Reset measurement is performed.
Set/Reset process
In the Set/Reset process of this device, the + voltage sweep is the Set process, and the – voltage sweep is the Reset process.
There is a part during the Set process where the resistance suddenly decreases, and the voltage at this time is called Set V. From this point on, a filament is formed inside the device, and the device changes from HRS to LRS.
There is a point during the reset process where the resistance begins to increase. The voltage at this point is called Reset V. From this point on, the filament inside the device breaks, changing from LRS to HRS.
Knowing the value of Set/Reset V through DC sweep allows us to set the conditions of the Pulse to be used in the next test, the Pulse test.
PD test
PD test is also called pulse test because it operates the device with pulses. As the name suggests, it is a method of measuring by applying pulses to the device rather than DC sweep.
Let's understand by looking at the measurement method and measurement results.
PD test method
If you look to the left, you can see the PD test scheme. Both the potentiation/depression process consists of a write pulse and a read pulse. During the write process, a set/reset pulse is applied to change the state of the device, and the resistance change of the device is checked through the read process. Therefore, the read pulse uses a voltage that is low enough not to affect the device. One thing to keep in mind is that the device resistance changes depending on the read voltage, so you must always read at a constant voltage and specify it. I measured at 0.1 V.
On the right are the PD test results. The x-axis is the # of pulses, or the number of pulses applied. The y-axis is the conductance, or 1/resistance. Through the potentiation process, the device is set, and you can see that the resistance decreases and the conductance increases accordingly. Next, through depression, the device that was set is reset, and you can see that the resistance increases and the conductance decreases. How can we know the performance of the device from these results?
PD test results and analysis
In the PD test, we need to obtain the resistance of the device, the ratio, and finally the nonlinearity.
First, the reason why the resistance of the device is important is because it determines the energy consumption of the device. P = V * I * t, right?? If the set pulse is 2V and 100ns and the resistance of the device is 1MΩ, then I = 2/1M = 2uA.
P = V * I * t = 2V * 2uA * 100ns = 0.4pJ
It has extremely low power consumption, right? So one of the advantages of ReRAM is its energy efficiency.
Second, you need to know the On/Off ratio. The higher the ratio, the better, but it's not just a large ratio; it also needs to have a large number of pulses. This means that the resistance shouldn't change all at once. This increases the number of states that this device can distinguish, and that's how good its analog characteristics are. (I'm not sure if this explanation is correct;; If you don't understand, please contact us by email!!)
Finally, we need to obtain a value called nonlinearity, which can be calculated by fitting the following formula.
Gmax and Gmin represent the maximum and minimum conductance values in the PD test result, respectively. The resistance is varied through pulses, and the more linear this change is, the better the analog characteristics, right? Nonlinearity refers to how much the PD test result deviates from a linear trend. The closer it is to 0, the better the analog characteristics of the device.
To summarize, a good analog RRAM is one that consumes less energy, can represent a large number of resistance states, and has linear resistance changes.
So why should we create these devices with such excellent analog characteristics? What applications are we researching for RRAM? To put it simply, these analog devices can be used in AI, such as deep neural networks (DNNs).
Retention test
RRAM is a non-volatile memory used as long-term memory (LTM) (they're the same thing). Then, after setting the device, it must be maintained in the low-resistance state (LRS). Therefore, after setting, a test is performed to monitor the device's resistance, which is called a retention test. The test result plots time on the x-axis and resistance (or conductance or current) on the y-axis. Of course, the measurement must be made at the same read voltage, and the voltage must be low enough not to affect the device. (I measured it at 0.1 V)
If you look at the graph above, you can compare the retention test results of STM (short-term memory) and LTM. You can clearly see that STM returns to its initial resistance level quickly after being set, but LTM maintains that resistance level after being set. This test is measured while varying the temperature, but I forgot the temperature environment in which it is usually measured. If anyone knows, please let me know by email. (I think it was 85 degrees, but my memory is hazy;;)
Endurance test
When operating a device in a real-world environment, it's not just a single operation, but rather multiple, repeated operations. Therefore, repeated testing under the same measurement conditions is required, ensuring consistent results each time. This is called endurance testing. Both DC sweep and PD testing are endurance tests, and most tests run for 100 to 1,000 cycles. Conditions are set and the test is run overnight or for several days.
The DC sweep endurance test is performed over 100 cycles. While the DC sweep results remain constant even after repeated measurements, the PD test results appear to be changing. A good component requires consistent test results.
ETC
variation
For high yield, variation must be small. Cycle-to-cycle variation (C2C) and device-to-device variation (D2D) are checked to determine the device yield.
STDP(Spike-timing-dependent plasticity)
Finally, we need the STDP test results. I remember that this test concluded by confirming that the measurement results follow the general STDP trend rather than measuring the performance of the device, but I don't remember clearly;;;; If I get a chance, I'll try to organize a related article separately.
References: https://ietresearch.onlinelibrary.wiley.com/doi/10.1049/iet-cds.2018.5388
