Advertisement

Analog Angle Blog

Point/Counterpoint: Let’s Debate Integration

You undoubtedly know that 3D printing — also known as additive manufacturing (AM) or rapid prototyping (RP) — is a hot topic these days. The publicity and hype almost seem that whatever your prototyping or manufacturing problem may be, 3D printing is the answer.

In some ways, this attention is well justified; just read this amazing story of how it was used to create an implantable, bio-resorbable temporary tracheal splint for an infant: “Baby breathes easy with laser-printed air tube.”

At the same time, we all know that sometimes the hype gets ahead of the reality. That's why I especially enjoyed an article in a recent issue of Desktop Engineering: “Point-Counterpoint: Additive vs. Subtractive Rapid Prototyping.”

In the piece, proponents of AM made the case for their approach, while proponents of more traditional manufacturing techniques, such as milling and drilling, did the same for theirs. Of course, the reality is that neither technique is the right or best one in all situations; the correct answer depends on many factors, including material to be used, unit volume, time to market, and cost, just to cite a few. As usual, it's largely about the tradeoffs and constraints.

It's somewhat the same in the IC world. Integrating “more” onto a single die has been the unstoppable trend since the first days of the IC, where the “more” can encompass a wide range of attributes: more functionality, more memory, more peripherals, more buffers, more I/O, more of whatever the end-application can use. Not only does this make a lot of sense, it makes a lot of products possible, no doubt of that. Without bigger ICs with more of everything and anything, many of the devices we take for granted, ranging from small smartphones to large computers, would be impossible to build with the features and price we want to meet.

Still, there are many times when integration isn't the answer. For example, if application requires a precise analog front-end amplifier or signal conditioner, designers often go with a single-function, basic IC having few or no other features beyond its primary one. That's why you see new products being introduced by the analog vendors such as op amps with ultra-low bias current for sensor inputs.

Yet sometimes, the push to use an IC with “everything” integrated onto it, and avoiding that single-function, high-performing analog component, is pretty strong. Management may insist that a more highly integrated part is the better choice, and sometimes, they are right: Perhaps the deficiencies — if any — of the integrated part can be overcome by better layout, lower-noise supply rails, some clever algorithms, or a more elaborate calibration approach, thus simplifying the BOM and lowering the cost. As in most engineering situations, the correct answer is, “It depends.”

That's why I'd like to see a one-on-one debate or perhaps a panel with IC designers, circuit designers, and other knowledgeable design experts doing a “point-counterpoint” on this topic, and real case histories as examples of what worked when, and also what didn't work out. I am fairly certain it would be both interesting and lively.

Have you ever been involved in this sort of technical dispute? How was it resolved? What was the outcome?

Related posts:

7 comments on “Point/Counterpoint: Let’s Debate Integration

  1. RedDerek
    September 15, 2013

    I have no issues with lots of integration for MY projects, but the cost would clearly not justify the means. Also, it comes down to volume of the product and cost. A highly integrated IC would end up being very application specific that the market is very limited – thus low production volume needed.

  2. BillWM
    September 16, 2013

    Precision is as you say difficult with high integration components — with that said I've done V&V for Single chip ADSL modems, VOIP phones, and just a few chips for Wireless devices —  Volume of the units sold is often key when integrating, but new FPGA and MCU's can often hit 10-14 bits of resolution which is “Good Enough” for many applications — and with just a few parts hit 24-32bits  which allows for things like Amazing Analog, Complex Digital, and Complex SW for example to occupy a very small space, or previously RF functions to become part Firmware.

  3. Nick Gray
    September 16, 2013

    There are approaches to programmable analog that would preclude the application specific limitation. However, those approaches have their own limitations. For example, analog switches are used to make connections. These switches do have a resistance that varies with process, temperature and voltage levels, meaning that there are variations in performance with these. Also, there are a lot of switches needed for high versatility, taking up die space. Minimizing the size of these switches will reduce die size at the cost of higher switch resistance in the on state and higher variation in resistance with temperature and process. Nonetheless, careful design procedures can minimize the disadvantages.

  4. Brad_Albing
    September 22, 2013

    @Nick – yep, another example of the almost endless balancing act you need to do to work thru a design: size, on-resistance, off-leakage, speed, breakdown voltage… round and round we go.

  5. Steve Taranovich
    September 29, 2013

    @RedDerek—I imagine you have a tough decision to make regarding critical time to market with one of your products vs. cost of a highly integrated IC. There are some AFEs out there, for example, at pretty relative low cost. Considering design time and cost to properly put a few separate ICs together in a design with good layout, grounding, bypassing, etc. for each of them, I would think that the integrated part might be pretty close to the few separate IC total cost.

  6. RedDerek
    September 29, 2013

    @SteveT – I make the decision about integration quite often when doing my designs. I have one design right now that has its challenges. To buy old stuff, or use a new microcontroller. Tough part is decoding the EPROM hex data to an assembly code to figure out what the processor is doing. Then program up a microcontroller which is capable of replacing the current processor, SRAM, EPROM, UART and PIA.

  7. Steve Taranovich
    September 29, 2013

    @RedDerek—that's probably one of the toughest, most time consuming jobs in moving to as new microcontroller. Hopefully you can use the same development tools too.

Leave a Reply

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