Analog design engineers do not frequently come across JFET (Junction Field Effect Transistor) or triode vacuum tubes nowadays. But, when designing an integrated circuit in a standard CMOS process, they may use native MOSFET (Metal Oxide Semiconductor Field Effect Transistor) devices from time to time. What could the common characteristic be of a 12AX7 (vacuum triode), a MMBFJ201 (JFET) and a 65nm native transistor deeply integrated in an IC, other than the fact that all three can probably be found in my garden shed?
Those devices all have three main terminations, presented in figure 1, if we exclude the substrate pins of the transistors and the heating filament of the vacuum triode. One terminal can be seen as a control terminal, the grid of the 12AX7, the gates of the MMBFJ201 and the 65nm native MOSFET transistors. In most configurations, the voltage between the grid and the cathode of the 12AX7 controls the current flowing between the plate and the cathode of the tube. Similarly, the voltage between the gate and the source of the transistors controls their drain-source current. Increasing the grid or gate voltage increases the plate and drain currents, respectively.
Figure 1: Electrical symbols of a vacuum triode, an N-type JFET, a native MOSFET, and an enhancement-mode N-type transistor.
This voltage control to output current characteristics is what forms the transconductance of the transistor. By adding a load to the plate or drain of the device, one creates a simple amplifier. This is where a small variation of the grid or gate voltage yields a larger variation of the plate or drain voltage. So far this is all very similar to the operation of a standard enhancement-mode planar transistor, and remotely like a standard bipolar transistor. From an analog designer's perspective, the main difference between those three devices and the standard enhancement-mode transistor, as shown in figure 1(d), is the value of the plate or drain current when the control voltage is zero. The enhancement-mode transistor commonly found in integrated circuits are normally OFF devices. That is when the gate-source voltage of the transistor is zero, there is practically no drain-source current through the device. On the contrary, the vacuum triode, JFET, and native MOSFET are normally ON devices. For instance, when the grid-cathode or gate-source are zero, there is a plate-cathode or drain-source current flowing through the device. A negative grid or gate voltage is needed to turn OFF the device. Normally ON devices can be difficult to use in many situations, but this characteristic can be extremely handy when designing simple always-ON bias circuits. Looking at the simplified schematic of the TL081 shows an example of such use. A JFET with the gate connected to the source of the transistor gives a simple and cost-effective current source. This small current can then be used to generate a precise voltage reference or current.
M0N0, the ultra-low power microcontroller from Arm Research, takes a similar approach for continuously monitoring the battery voltage of the system. A simple, low power, cost effective yet accurate enough approach for continuously monitoring a battery voltage.
M0N0 has been designed to be part of a sensor system that can be deployed in dangerous or difficult to access environments. Built for a single cell battery supply with a voltage ranging between 1.5V at full charge down to 1V when discharged, it embeds an Arm Cortex M33™ CPU with DSP and SIMD extensions, 128kB of ROM and 16kB of SRAM. What makes M0N0 unique is that despite those powerful features, it not only draws less than 10nW from the battery when in shutdown and retaining 4kB of data, its active power can be as low as 10uW (0.8MHz clock frequency) and its computing speed as high as 38MHz (3mW active power). We demonstrate its energy-efficient operation with a keyword spotting application running out of a single cell alkaline battery. All batteries discharge with time, and replacing a sensor or just its battery when necessary is less dangerous and more cost-effective than a more frequent periodic checks and/or replacements. By continuously monitoring the voltage of the battery supplying M0N0, one can detect a near-discharged battery, automatically trigger a sensor response and send an alarm message upstream to the data gathering system.
M0N0 detects a discharged battery by comparing the battery voltage against an arbitrary reference. If the battery voltage drops below the reference, the battery is likely to be close to being discharged and an alarm is raised. It might sound simple, but doing this while dissipating less than half a nW of continuous power is where the challenge is. At such power levels, less is often more. The simpler the circuit the better. Here, the normally ON property of the native N-transistor available in the 65nm process we used for M0N0 comes in handy. By connecting its gate and source to the ground, it gives a simple way of generating a constant current. By adding a diode-connected transistor between the source and the ground of the native fet, one creates a self-biased voltage reference, as presented in figure 2.
Figure 2: Self-biased voltage reference.
This well-known circuit is detailed with some variations in [1], [2] and [3]. Simply, if we add an initially discharged capacitor across the diode transistor, M1, the source voltage of the native transistor, Vref, will rise until the native transistor characteristic intersects the diode-connected transistor characteristic, as presented in figure 3. If the voltage Vref is too low, the diode characteristic implies that Idiode is less than Ifet, therefore charges accumulate on the reference node and Vref rises, leading to two effects. First, the diode characteristic implies that Idiode increases with a Vref increase. Second, the native characteristic implies that Ifet decreases with a Vref increase as the gate of the transistor becomes even more negative relative to its source. Eventually, as Kirchhoff’s Current Law states, Ifet = Idiode = Ib and the reference voltage, Vref, is stable.
Figure 3: Current-voltage operation characteristics of the diode-connected transistor M1 and drain current as a function of the source voltage of the native transistor M0.
Designing references is all about stability across temperature and supply voltage variations. By using specific size ratios for the length and width of the devices between the native transistor and the diode connected transistor, the reference voltage Vref can be stable over the wide temperature range. While M0N0 is designed to operate between 0°C and 85°C, its battery monitor goes far beyond that range. Supply voltage variation can be dealt with by using a long length native transistor, a simple analog design method. We now have a simple, low-power voltage reference that gives a constant voltage of 313mV. It is typical power dissipation is near 50 picowatt at room temperature - accurate enough for our targeted application. One could divide the battery voltage using a resistive divider or a diode ladder, and use a simple voltage comparator as presented in figure 3. However, this is not simple enough for Arm researchers. Large resistances for R0 and R1 would be needed for low current, making the circuit expensive in silicon area. Also, the power consumption of the comparator would cancel most of the benefits of using the two-transistor voltage reference shown in figure 2.
Figure 4: Straightforward approach of the architecture of a battery voltage measurement circuit.
We kept it simple by flipping the problem around, comparing currents instead of voltages. The normally ON device generates a current. By sizing the native transistor and the diode connected transistor, we made sure that this current is the current needed to bias the diode-mounted transistor, M1, at a fixed voltage irrespective of its temperature. This is a much more powerful tool than a fixed reference voltage. We then mirrored that current, and compared it against the current flowing through a diode ladder of N diodes biased by the battery voltage as presented in figure 4.
Figure 5: Diode ladder.
The voltage across each diode is the battery voltage divided by the number of diodes. We have designed our circuit making sure that all diodes are the same, therefore the current IL equals the reference current Ib when:
By copying the bias current Ib and the ladder current IL using simple current mirrors as presented in Figure 5, we get the comparison as a voltage at the node out. If the battery voltage is higher than N times Vref, the ladder current IL is larger than the bias current Ib and the output node is fully discharged to the ground. If the battery voltage drops, as in the case of a discharged battery, the voltage across each diode of the ladder becomes less than the reference voltage. The ladder current I is therefore lower than the reference current Ib, charges accumulate on the output node, and the output voltage rises. This then in turn raises an alarm that can be used to trigger M0N0’s response to a near-full discharge of the battery.
Figure 6: Simplified schematic of the battery voltage measurement circuit.
Building on the principle of the circuit in figure 6, we have added a hysteresis cycle to the under-voltage alarm (UVLO) as well as an over-voltage alarm (OVLO). This is to protect M0N0 should anything happen to the battery supply, as all of this is operating at 100 pW at room temperature and less than 2 nW at 85°C. Figure 7 presents those alarm thresholds over temperature for an untrimmed circuit, straight out of the box.
Figure 7: Experimental measurements of a battery monitoring thresholds.
With only a handful of constantly biased branches, we have designed a circuit that draws a fraction of a nW in typical conditions and avoids unnecessary maintenance once out in the field. M0N0 is advertised at 10nW shutdown power, but it is not all sleeping power. Under the hood there is a little piece of circuit that is constantly monitoring your battery, making sure there is just enough energy stored for M0N0 to perform at the next wake-up request.
Why did we mention the 12AX7 at the beginning of this blog post then? It shares its normally-ON property with the native-transistor deeply embedded in M0N0. We have designed a circuit with this latter device as generations of engineers before us have designed with valve triodes and JFET transistors tailoring the bias point of the device to get the most of them. That is why the legend says that this circuit idea comes from looking at push-pull valve amplifier schematic.
Questions? Contact Benoit Labbe
What else is there to know about M0N0? Explore our research so far using the following links.M0N0: An Arm Research platform for N-ZERO sensors
M0N0: The devil is in the details