As part of a recent project here at Microncubed we had the opportunity to tear-down and look inside a silicon MEMS flow sensor. Flusso, a company based in Cambridge, U.K. developed this compact sensor in order to measure air and gas flow. A team experienced with microscale thermal devices founded Flusso in 2016. And in 2022, a Shanghai based private equity firm purchased the company for £28M. Today, there are two products, a mass flow sensor and an air velocity sensor. Here, we take a look at the FLS110, the previous generation of the mass flow sensor, which has the same form factor as the more recent FLS112.
What does the package look like? The device is encapsulated in a 6-pin DFN package which has a 3.5 × 3.5 mm footprint. It is 3 mm in overall height and includes input and output ports for the flow. The datasheet tells us that the device can be used either in a through-flow or bypass configuration. In a through-flow setup, it is possible to measure flow rates in the range 4 to 200 standard centimetres cubed per minute (sccm). The upper limit is extended in the bypass configuration. The package itself contains no digital circuitry, only the analog components required for the flow measurement. The user has to connect the device to both digital and analog pins of a nearby microcontroller.
Opening the package
Let’s take a look inside the package. We solder it onto a PCB and use a scalpel blade to detach the lid. The micrographs show the package before and after lid removal. Before lid removal, the input and output flow ports are visible. Afterwards, we can see the silicon die inside the package with 6 connecting wire bonds. The dimensions of the die are 1.5 x 1.2 mm. At this low magnification, the most notable feature on the die is the transparent central window. This is a dielectric membrane of dimensions 750 x 750 um. What’s the purpose of the membrane? The heating element is insulated from the silicon substrate, since a thermal technique measures flow. Silicon is an excellent thermal conductor. The approach used here is to support that heater on a thin dielectric membrane.
On the datasheet we see that pins 1 and 6 are not intended to be connected (DNC). This leaves one pair of pins (2 & 5) labelled FS which attach to the flow sensor. And the remaining pair of pins (3 & 4) labelled TS belong to a temperature sensor. So, the DFN package only contains two analog components, the flow sensor and the temperature sensor.
The MEMS die in more detail
In order to show more detail, we image the die at higher magnification. Now we can see that the flow sensor and the temperature sensor are both high aspect ratio tungsten wires. Flusso use the acronym T-MEMS for the manufacturing process, where T is for tungsten. Both wires appear to share the same geometry. They differ in that the flow sensor is on the dielectric membrane while the temperature sensor is on the silicon die. In addition, the DNC pads connect to a wire that goes all the way around the perimeter of the membrane. Let’s assume that it provides an electrical check of the membrane’s integrity. If the membrane breaks, the wire is likely to break too.
There are unbonded voltage probes (far left bond pads) on the temperature sensor (i.e. excluding the bond pad region). These allow the behaviour of the central region of the temperature sensor to be probed. They were likely used for development, or during an in-line test in manufacturing. The zoom-in shows the connection of these voltage probes in more detail.
Measuring the dimensions of the temperature and flow sensor wires, we find them to be 580 um long and 2 um wide i.e. having an aspect ratio of ≈ 300. The high aspect ratio being necessary to form a wire of sufficiently high electrical resistance.
A dielectric layer separates two layers of tungsten metallisation. We can see this from the crossing of the membrane integrity loop by the flow sensor wire. Also seen is a further metal layer, in copper, on the bond-pads and for the dummy structures on the right and left of the die.
The flow-sensing methodology for this device is thermal. We believe it works in the following way. The temperature sensor, via its electrical resistance, measures temperature at the upstream side of the die. A current dissipates electrical power in the flow sensor. This causes a temperature rise that depends on the applied power and flow rate. This temperature rise, measured from the flow sensor’s electrical resistance, determines the rate of flow.
Temperature sensing with the tungsten wires
In order to make a temperature measurement using both the flow and temperature sensors, we need to determine their resistance and temperature coefficients. Placing the chip in an environmental chamber, we measure the resistances over a range of temperatures. First, we find the resistance of the flow sensor (131.6 Ω), temperature sensor (132.0 Ω) and membrane integrity loop (204.2 Ω) at 40 C. Then we measure the increasing resistance with temperature up to 100 C. To first order, the resistance varies with temperature as follows: R(T) = R0(1 + ⍺ΔT). In the expression, ⍺ is the temperature coefficient of resistance, ΔT is the temperature rise and T0 is the initial temperature (in this case 40 C). Fitting to the experimental data (see figure), we find temperature coefficients of resistance for the flow (0.00233/C) and temperature sensors (0.00234/C).
The nearly identical resistances and temperature coefficients suggest that the geometry and material for the temperature and flow sensors is the same. And the temperature coefficient for the membrane integrity loop is similar (0.00220/C). Note that these temperature coefficients are approximately half that for bulk, elemental tungsten. Such a reduction occurs in a thin film where the finite thickness and film morphology both introduce additional electron scattering mechanisms, thereby reducing temperature coefficient.
Thermal resistance on and off the membrane
Thermal resistance (Rth) is the constant of proportionality between temperature rise (ΔT) and applied power (P) i.e. ΔT = Rth P. We measure the electrical resistance at different applied power in order to determine the thermal resistance of the temperature and flow sensors. And from the temperature dependent electrical resistance we determine the wire temperature as a function of this applied power (see figure). We observe a linear response, the gradient being the thermal resistance.
The difference between the thermal resistance of the membrane situated flow sensor (12.7 C/mW) and die situated temperature sensor (0.5 C/mW) is striking. The higher the thermal resistance of the flow-sensor, the more sensitive it will be. The temperature sensor has a low thermal resistance, so that power dissipated during measurement doesn’t cause significant self-heating.
It would be interesting to build a model of the flow-rate sensitivity of this device based on the thermal parameters and geometry, however that is outside the scope of this short post.
Conclusions
We looked inside the package of the Flusso mass flow sensor, FLS110. Essentially it consists of two high aspect ratio tungsten wires which act at thermometers through their temperature coefficient of resistance. One, on the upstream side of the die has a low thermal resistance and determines the die temperature. The other is located in the middle of the die on a dielectric membrane and has a relatively high thermal resistance. Its temperature increase for a given power dissipated depends on the flow rate, enabling functionality as a flow sensor. We measured two parameters critical for the sensitivity of the device as a flow sensor: the temperature coefficient of resistance and the thermal resistance.
The concept of a micro fabricated silicon hot-wire flow-sensor has been around for many years. A nice reference is van Oudheusden’s Ph.D. thesis, “Integrated silicon flow sensors”, from 1989. It appears that Flusso have achieved a nice implementation of a silicon MEMS flow sensor, which uses modern MEMS manufacturing, is compact and conveniently packaged.