Advertisement

Blog

Highly Integrated ICs: Who’s Running the Show?

In the face of the constant pressure from customers and managers to reduce the overall size and cost of mixed-signal designs, one typical solution has been to shrink the design size and lower the number of IC components through integration.

By taking functions that were previously implemented piece-by-piece using discrete components and merging them into the central controlling core of the design (often a microcontroller), the design is made smaller and simpler, and costs are reduced. This is now taken as such conventional wisdom that it seems almost too obvious to be explaining it.

However, as with most engineering decisions, there are potential trade-offs as well as benefits when moving to a more tightly integrated design. The benefits seem obvious — your design will have fewer ICs and supporting discrete components, and it will take up less PCB area and require fewer interconnects. Therefore, overall cost and power consumption of the design will go down. A side benefit is often that a simpler design (simpler at the PCB schematic level, at least) is easier to develop; easier to maintain; easier to modify; and with fewer active components, easier to test before putting into production.

But what are some potential tradeoffs of this approach? Many of them have to do with control. First, consider parts sourcing. If you develop a system with a “generic” central microcontroller, and the analog functions live in separate devices, the components of the system are much easier to swap out or second source if needed. If the microcontroller does not directly incorporate the analog functions, it can be traded out in the design for any microcontroller of equivalent capability that is capable of communicating with and controlling the analog slave devices in the system. Similarly, if some or all of the analog devices are separate, they can be replaced with equivalent components if a newer, better IC becomes available. But if the microcontroller is instead a special-purpose one, with integrated analog components, it may not be possible to find another microcontroller from a different manufacturer that does exactly the same thing — you may be locked into one device for the purposes of your design.

Similarly, if you spread the functions out, you are in control of exactly what your system is capable of and what it is not. Moving to an integrated “all-in-one” microcontroller means that you are limited to the combinations of features that are available for that device. If you need three ADCs, and the microcontroller you are considering only allows two, you will have to add another external ADC to the design or change to a different microcontroller. And if there are components to the microcontroller that support capabilities or interfaces that your design does not actually use, then those capabilities (and the additional cost and complexity in the microcontroller that they represent) are wasted in your system.

In the best partnerships between mixed-signal microcontrollers and their customers, these concerns are reduced by developing specifications for new devices as a joint effort. By taking a customer’s specific needs into account when creating a new device, it is possible to tailor an IC design specifically to what a customer requires to go from one generation to the next of their system, while reaping the maximum rewards from cost and power reduction. But this requires a committed joint effort up front, and a solid working relationship. As with so many other things in engineering, communication, early and often, is the key to ending up with the best possible design for your product.

6 comments on “Highly Integrated ICs: Who’s Running the Show?

  1. amrutah
    February 16, 2013

    @Aaron,

    “Moving to an integrated “all-in-one” microcontroller means that you are limited to the combinations of features…”

      By scaling technology we may even integrate a few more modules to make it generic, but in order to support more applications no. of pins will increase, involves rigorous testing (hence increase of cost)…

      We will also have to consider the EM constraints when the feature size gets reduced…

  2. patrick_m
    February 16, 2013

    Hi @Aron, nice post! I see so much great innovation happening at the integration and power-management level, that when you say.

    ” And if there are components to the microcontroller that support capabilities or interfaces that your design does not actually use, then those capabilities (and the additional cost and complexity in the microcontroller that they represent) are wasted in your system.”

    I have to wonder if these extraneous parts, and their associated cost/power losses/failure modes are less a concern now than in the past, to the point that it's almost a non-factor for most integrated analog designs. Only corner-case, totally optimized-in-one-direction (power? cost?) designs would consider it an issue, in the face of the benefits of integration.

    Anyway, thanks for the overview, nicely done.

  3. eafpres
    February 16, 2013

    Hi Aaron–you sparked a couple thoughts in my head reading your nice article.  First, for any integrated part, do you think that the future risk of the part going EOL increases or decreaes from the past?  In the past, lower functional ICs and even passives could go EOL, requiring a re-design or at least re-verification to ensure a “drop in” replacement would work.  With an integrated part, a drop-in will be less likely, but do you think these parts will have longer life cycles from the manufactuers?

    The other question for you is what fraction of highly integrated parts come out of a specific customer need that is then “productized” and put into the general market, vs. parts designed by manufacturers based on their assessment of market needs?  I wonder if in the first case parts perform in certain ways that are less obvious when the generic data sheets are published and others than the original customer use them?

  4. Aaron Minor
    February 25, 2013

    First, for any integrated part, do you think that the future risk of the part going EOL increases or decreaes from the past?  In the past, lower functional ICs and even passives could go EOL, requiring a re-design or at least re-verification to ensure a “drop in” replacement would work.  With an integrated part, a drop-in will be less likely, but do you think these parts will have longer life cycles from the manufactuers?

    Just my personal opinion, as I am not directly involved in decisions about moving parts to a 'not recommended for new designs' sort of status…

    I think the life cycle of any part will depend somewhat on how many customers are using it.  If it is a part that is tailored very specifically for one customer's needs, and the customer is still actively using it, that customer would naturally be consulted before developing a follow-on or replacement for that part.  But this part may have a shorter life cycle, because if the 'big customer' agrees that a change is ok, and that customer is buying 95% of the stock, then smaller customers that only buy a few may not be as involved in the decision making about the next version of the part.

    If the part is more of a 'catalog design' or a more generic device that could conceivably be used for a number of different applications, then the life cycle will probably be longer.  Medical and industrial applications in particular tend to have long life spans, as devices in this category do not typically get replaced as frequently as consumer electronics.

    The other question for you is what fraction of highly integrated parts come out of a specific customer need that is then “productized” and put into the general market, vs. parts designed by manufacturers based on their assessment of market needs?  I wonder if in the first case parts perform in certain ways that are less obvious when the generic data sheets are published and others than the original customer use them?

    Hard to say; this is more of a company strategy question than a technology one, so this will vary from one IC manufacturer to the next.  Some products are produced largely to fill a specific customer's need; other products are produced with a certain kind of application or technology in mind, with the idea that many different customers would be able to use it.  Still other products are created to target existing products/features of parts that are already in the market, with the idea of replacing competitors' devices in existing design sockets.  If a device is popular enough, other companies will often attempt to produce their own pin-compatible version, if they think they can do the same thing but at less cost, with less power, etc.

     

  5. Aaron Minor
    February 25, 2013

    I have to wonder if these extraneous parts, and their associated cost/power losses/failure modes are less a concern now than in the past, to the point that it's almost a non-factor for most integrated analog designs. Only corner-case, totally optimized-in-one-direction (power? cost?) designs would consider it an issue, in the face of the benefits of integration.

    It's a matter of degree.  If 90% of what is on the part meets a designer's needs, and the portion of the features that are not used by the application does not consume an inordinate amount of power, real estate or pins, the designer is more likely to say, 'well that's close enough for now at least'.

    If half the features on a part are irrelevant to what the designer is trying to do, they are more likely to look elsewhere.

    Large customers also have more flexibility in this regard, as they pull enough weight purchasing-wise to be able to say, “well we like part A, and we will use it for prototyping, but we would like a variation of this called A2, which takes out this thing we don't need, and adds a few things here and over there…”, and thereby get the IC manufacturer to optimize an existing part for their particular design.

    This represents much less of a committment for both sides as the customer can prove to themselves that their application will work with the existing part before ordering the new variant, and the manufacturer doesn't have to develop a part from scratch, just tweak one they already have.

     

  6. Aaron Minor
    February 25, 2013

    By scaling technology we may even integrate a few more modules to make it generic, but in order to support more applications no. of pins will increase, involves rigorous testing (hence increase of cost)…

    It's a balancing question; designing in potential flexibility is good up to a point, but the trick is in knowing what combinations of features customers are likely to want to use.

    It's easier to design (and test and manufacture) one device that can be configured to act 5 different ways (for example, with different package pinouts) than it is to actually produce five different ICs.  But the customized ICs may be smaller and/or consume less power, since features may be actually removed, not just shut off.  It depends on the characteristics of the features in question.  If a feature on a part doesn't consume much in the way of area, power, or pins, it doesn't make much sense to remove it from a design as long as it isn't actively preventing something else the customer wants to do.  Pins are often multiplexed when possible at the I/O pad level, to enable unwanted features to be shut off and ignored so that they do not take up pins that can be used for something else.

     

Leave a Reply