Advertisement

Blog

Filter Techniques in Circuit Integration

As a follow-up to my previous blog, Integration Choices: Analog Filters vs. Digital Filters, the next step would be to discuss how to integrate filters into the IC itself.

What techniques and architectures have evolved over the years that make integrating a filter onto an IC easier, smaller, better matched, variable, and lower power than ever before? Let's take a look at some of the most innovative ones, and I invite you to comment and add your own favorites too.

The MEMS resonator
In wireless communications circuitry, band-pass filters (BPFs) are widely used and need pretty high performance in order to minimize signal loss and attenuation of out-of-band signals and interferers. Let's take a look at MEMS resonators (See Figure 1) as an alternative to the traditional quartz crystal and surface acoustical wave (SAW) filters.

Figure 1

The MEMS resonator may be a good, low-cost alternative to conventional filter components. The square in the center of the device is suspended and held in place by the four arms. The device mechanically resonates at particular, well-separated frequencies. (Source: Reference 1)

The MEMS resonator may be a good, low-cost alternative to conventional filter components. The square in the center of the device is suspended and held in place by the four arms. The device mechanically resonates at particular, well-separated frequencies.
(Source: Reference 1)

The BPF usually resides in the RF and IF stages of the RF analog front end (AFE). A major drawback in the MEMS resonator has been high series motion resistance to the signal that will lead to poor insertion loss. There are a number of techniques to help improve the attenuation of the signal due to this technique, but one effective approach involves reducing the gap between the electrodes and the resonating body. This technique reduces the motional impedance with the fourth power of the transduction gap. See Reference 1 for more details.

Using post-processing techniques, the basic CMOS process remains untouched. New devices are fabricated on top of the initially processed die, leading to a miniaturized and low-cost, mass production technique.

The active inductor[2]
CMOS in RF has become the go-to process for low-power and mixed-signal integration in the industry. Let's look at an example of an RF band-pass filter (BPF) for a receiver front-end and see how an active inductor can play a critical role by implementing it in the CMOS process.

It has been demonstrated that an on-chip and tunable BPF can be constructed with an active inductor architecture in CMOS. The traditional architecture of an RF receiver front end has been to have the band-pass filter external from the CMOS receiver using discrete components such as in a lumped LC or a SAW filter design. With the advent of the software defined radio (SDR), an on-chip filter that can tune across a wide frequency range and also has adequate selectivity, is imperative to the high integration goal and ultimately single-chip solution of the SDR.

Until recently, passive spiral inductors have been integrated into the CMOS IC for this function. Their disadvantage is ohmic and eddy current losses and sensitivity to EMI. Shields can be added to help this, but are added cost in assembly and bill of materials (BOM) costs in the system. Spiral inductors can only achieve low Q-factors and low inductance values, plus they take up a great deal of real-estate on the chip when multiple inductors are needed. Enter the active inductor architecture.

The active inductor is architected using the gyrator principle shown in Figure 2(a). Figure 2(b) shows how the current-voltage characteristic of the intrinsic capacitance in the transistors is inverted to make the load appear inductive in nature. See Reference 2 for more details.

Figure 2

(a) The gyrator topology creates an inductive load Zin. (b) The equivalent passive circuit is shown for the realized circuit. (Source: Reference 2)

(a) The gyrator topology creates an inductive load Zin .
(b) The equivalent passive circuit is shown for the realized circuit.
(Source: Reference 2)

At the IC level, the inductor is realized by combining a common-source transductor (-gm ) with either common-gate or common-drain transconductors (+gm ). See Figure 3 for some examples of various popular configurations.

Figure 3

There are a few good active inductor configurations shown in literature, as seen here. (Source: Reference 2)

There are a few good active inductor configurations shown in literature, as seen here.
(Source: Reference 2)

This architecture has the benefits of small area on the chip as well as large inductance values. The center frequency (ω) and Q of this active inductor tunable filter has the added advantage of being tuned by varying the transconductance (or capacitance) of the transistor.

Reference 2 has more details regarding improvement of the Q for this type of filter.

There are still a number of issues with this technology, but Reference 2 outlines some ways to help solve these.

Moore's law implies that as RFIC semiconductors continue to downscale, the complexity of the ICs will increase.

Active-C architecture[3]
In this architecture, we will look at a third order voltage mode active-C band pass filter as an example of the Active-C technique. This architecture only consists of three equal value capacitors, a single dual output current controlled current conveyor (DOCCCII), Figure 4, and two current controlled current conveyors (CCCIIs). See Reference 3 for more details.

Figure 4

The block diagram of the DOCCCII. (Source: Reference 3)

The block diagram of the DOCCCII.
(Source: Reference 3)

This architecture is well suited for IC design since it uses a minimum number of active analog building blocks as well as a minimum of passive components. All of the used capacitors are grounded and resistors are not required. Nice!

Figure 5

The transistor structure of the DOCCCII. (Source: Reference 3)

The transistor structure of the DOCCCII.
(Source: Reference 3)

Current conveyors have the unique properties of wide dynamic range, high signal bandwidth, small chip area and low power consumption making for an ideal architecture for today's CMOS integrated RFICs.

The parasitic resistance at Port X in Figure 4, is adjustable via the bias current, thereby allowing tunability. Biquadratic filters have been built using this architecture and even higher-order filters can be built by such techniques as cascading lower order filters, signal flow graphs, and state variable techniques. See Reference 3 for more details.

These are only a few of the elements and architectures that are enabling the growth highly integrated analog ICs. Please share with us any of the favorite designs and technolgies that you have used.

References:

  1. Mechanical resonators on CMOS for integrated passive band pass filters, S. N. R. Kazmi, C. Salm and J. Schmitz, MESA+ Institute for Nanotechnology, University of Twente, Enschede, Netherlands
  2. Development of Active Inductor in CMOS Tunable RF Bandpass Filter, Nur Diyana Mohd Asri, Norhayati Soin, Department of Electrical Engineering, University Malaya (UM), Kuala Lumpur, Malaysia.
  3. Realization of Active-C Voltage Mode Third Order Band Pass Filter with Current Controlled Current Conveyor (CCCII), Ashish Ranjan and Sajal K. Paul, Department of Electronics Engineering,Indian School of Mines, Dhanbad, India

Related posts:

10 comments on “Filter Techniques in Circuit Integration

  1. Scott Elder
    July 23, 2013

    Yannis Tsividis has published alot of work on MOSFET-C filters.  This is where the MOSFETs are tuned as a resistor by driving the gate voltage.  I built one many years ago for a multi-megahertz bandpass filter inside the read channel of a hard disk drive.

    Nowadays it seems that the objective is to move the signal into the digital domain as soon as possible (i.e. ADC) and solve the rest of the problem using DSP.  A simple anti-alias filter in front of the ADC is all that is required.  

     

  2. Steve Taranovich
    July 23, 2013

    Good reference Scott. I've seen Tsividis work in VLSI on the IEEE XPlore site for MOSFET-C filters—good stuff

  3. JeffL_#2
    July 24, 2013

    1. Your figure 2(a) seems to suggest that it emulates a grounded inductor. You mentioned that this could be used (among other things) to make an analog BPF. In many BPF implementations a substantial fraction of the inductors would NOT be in a grounded configuration (series LC section for ezample). Does a “differential” connection (effectively 2 Figure 2(a)s in tandem) work OK or would capacitive imbalances create problems? If there are problems is it necessary to change the filter topology or is there another configuration that will properly synthesize an ungrounded inductor?

    2. To roughly what RF frequency are these techniques considered practical and can you give instances where they have been put into production at those frequencies?

  4. Brad_Albing
    July 24, 2013

    @Scott – I assume this is much like the way we used to use JFETs as voltage variable resistors. Of course, they were depletion mode devices, so we needed a negative gate voltage. But otherwise, it seems similar – right down to the pesky problem of the need to characterize channel resitance vs. gate voltage so you could get predictable performance.

  5. Steve Taranovich
    July 24, 2013

    Hi Jeff—excellent questions:

    1. In a parallel circuit, the smaller quantity (reactance) has the greater effect on the total current. Think of the example of resistances in parallel, the smaller reactance draws more current and has a greater effect on the total Z. An option is to keep the “inductive” reactance high enough in the series RLC combinations, so that its effect would be negligible.

    However, in cascading these types of parallel filter architectures, it might be better to change the filter topology—you can still simulate an inductor in any other configuration. Check out the references at the end of this blog for more details.

    2. Actual implementaions show 600 MHz to 3.8 GHz tuning range with a noise figure of 5—see reference [1] at the end of the article.

  6. Netcrawl
    July 24, 2013

    Yes I read it wonderful stuff, its contained detailed information of VLSI works, the filters are being discussed effectively, good sources of reference. 

     

     

     

     

  7. kendallcp
    July 26, 2013

    Hi Jeff

    Floating inductors (and other higher order elements) have for decades been the bane of filter designers trying to migrate from tried and tested LC network design into something more active, whether for integration or just to avoid baseball-sized limps of ferromagnetic material.

    When asked (which is, for such an esoteric subject, quite infrequently), I usually counsel to avoid any of the active floating element techniques.  They present a lot of matching and common mode problems.  There are two main ways to achieve this, facets of the same bigger network synthesis picture:

    1  Use only grounded inductors – they are easy to synthesise.  Of course, chip designers like grounded capacitors too.  Something has to give way, especially if you've got series-resonant circuits…

    2  Transform your design so that a floating inductor becomes a floating resistor (or even a floating capacitor, that's rare though).  That's the basis of Bruton-transformed FDNR filters, one of the most elegant conceits in the entire active filter canon.  I wrote about this a while back:

    http://www.analog-eetimes.com/en/bruton-charisma-make-those-inductors-vanish-using-savvy-scaling.html?cmp_id=71&news_id=222901145

     

     

  8. Brad_Albing
    July 27, 2013

    @KCP – thanks – a good article to read for reference.

  9. SunitaT
    July 30, 2013

    In the design of frequency-selective integrated circuits, the lack of inductors is a serious drawback. In many apps, this drawback can be overcome by using conventional active-RC filter methods. In designing active-RC filters, one uses a combination of resistors, capacitors, and gain blocks in linear feedback loops to get the desired frequency selectivity without the need for inductors.

  10. Steve Taranovich
    July 31, 2013

    Another nice technique for integrated electronics—thanks SunitaT—the other advantage of active-RC filters is to decrease the value of the filter's capacitors

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.