Slashdot Mirror

← Back to Stories (view on slashdot.org)

Analog & Digital Chips On The Same Silicon

Posted by ryuzaki0 on from the put-'em-together dept.

jukal writes "Forbes.com writes: "Intel Corp. Monday announced plans to put some functions of analog and digital chips onto the same piece of silicon, its latest push into the communications semiconductor industry.", "which will be available early in 2004, could lead to a single-chip hand-held device that offers cellular phone, wireless-data-network and other connection services.", so, I quess this will be a competitor to the Texas Instruments' OMAP chip?"

13 of 83 comments (clear)

  1. Not new, it's called Mixed signal. by msgmonkey · · Score: 5, Informative

    A large section of embedded IC's have digital & analog on one chip. This has been done for years, just beacuse Intel are now doing it does n't make it news.

    1. Re:Not new, it's called Mixed signal. by jaoswald · · Score: 5, Informative

      You are completely right about mixed signal chips being a reality for a long time: you don't need to look any further than a video card in your computer to see that. Intel mentions "silicon radio" as if it is a new idea, but a company already exists called Cambridge Silicon Radio, so you can see it isn't just Intel in this business.

      I have a feeling that something important is being left out of this article. If you look at the original press release you see that it is a total mishmash of different Intel developments. The poor journalist was stuck trying to find a lead in this story (other than "Intel has bunches o' innovation") and zeroed in on the part that mentioned Moore's law, which he had heard before.

      The most interesting part that I see is the tunable laser using silicon photonics. Si has an indirect band-gap, which makes it not very good for making lasers and optical devices. That could be big news.

    2. Re:Not new, it's called Mixed signal. by Anonymous Coward · · Score: 1, Informative

      Mixed signal is not new, but SiGe on CMOS is new. This allows much higher speed analog circuits than traditional mixed signal devices. I believe that Intel is one of the first to commercialize this process. That's what makes this news.

    3. Re:Not new, it's called Mixed signal. by hagn · · Score: 2, Informative

      Its not only the classical "mixed signal", what they want to integrate onto a chip but also the HF components, like Low Noise Amplifier(LNAs), mixers and filters which work in the GHz frequency range. Thats also what they need the SiGe HBTs for. But i agree to you that this has been done before. See for example single chip bluetooth. On the other side this has not be done for GSM chips up to now, but Intel is not the only company working on that.

      --
      Marcus

  2. Mixed Analog and Digital isn't new at all. by Anonymous Coward · · Score: 1, Informative

    Single chip A and D have been available for at least 20 years. Hobbyists could buy single chip DVM kits and virtually ALL modems nowadays are Mixed A and D DSP chips.

  3. This is new? by Black+Cardinal · · Score: 2, Informative

    Maybe its new for this application, but hybrid analog/digital chips have been around for a long time. Anybody ever hear of an analog-to-digital converter, or perhaps a digital-to-analog converter?

    For that matter, inkjet printheads have quite a bit of both analog and digital circuitry on them, and they are made out of a single silicon die.

    1. Re:This is new? by jaoswald · · Score: 4, Informative

      An A-D converter is not necessarily a truly hybrid device. The point is that there are transistors that are good for producing gain, possibly at high frequencies. Those make up what are generically called "linear" chips. Mostly op-amps and so on.
      These tend to be bipolar junction transistors or related technologies. The key thing is that they tend to pass current all the time.

      Then, there are transistors which are good for switching, for creating logic gates & CPU logic. These tend to be CMOS field-effect transistors which are designed to only pass current when they are switching, in order to reduce power consumption so that you can raise the clock rate to obscene levels. However, logic gates are ideally non-linear: either on or off, with nothing in between.

      The problem is that these technologies are differently optimized, and aren't naturally compatible. Coming up with a process that can produce nice linear transistors along with high-performance logic gates is tough. You can also try to approach it from the other end: come up with some kind of circuit which can make nicer amplifiers out of lousy transistors.

      That's what makes true mixed-signal chips difficult: you either give up linear behavior, or increase current draw, or you give up the gate density and clock performance.

    2. Re:This is new? by pll178 · · Score: 2, Informative

      You're confusing the transistor devices with the circuit topology. You can build analog circuits from NMOS and PMOS transistors (CMOS process), in fact, when I took an analog circuits class at Berkeley, we only used NMOS and PMOS transistors, no bipolar transistors at all. We were able to build almost any analog circuit that we wanted. If you put a NMOS and PMOS in an inverter configuration to get a switch, but you can play some tricks to make the same transistors produce a current source or an op-amp.

      On a side note, in a CMOS process, you can create crappy NPN and PNP transistors called lateral NPN or lateral PNP. It has horrible gain, but if you needed to build a bandgap voltage source (or some other bipolar-like device), you can use this transistor.

  4. Problem with Mixed Signal Chips Like This by Anonymous Coward · · Score: 3, Informative

    As ambitious as this effort is, there are significant barriers to getting it to work properly.

    First off, analog and digital go hand in hand. All digital circuits are essentially analog circuits operating in a non-linear range. However, high-frequency analog circuitry is particularly problematic. Even basic structures such as phase locked loops and analog-to-digital converters can generate a lot of on-chip noise, both in the silicon substrate itself and through parasitic coupling above it. For basic PLLs, you need a good 50-100 microns of space between it and the nearest logic gate. Higher-speed cores will require structures like isolation tubs and additional spacing, and will significantly hamper placement and routing of the remaining circuitry. In other words, it is very easy to run out of die space and/or introduce signal integrity problems.

    Speaking of signal integrity problems, the smaller geometry ICs (0.18um feature size and below) are having their signal integrity problems get worse and worse. Noise, delay, and wire melt are common problems that need repair in the digital circuitry, and noise margins are getting razor thin as it is. Power distribution is also going to be a nightmare, considering that every analog block will need its own power, probably multiple FC lands per block. The thing is, the CAD tools aren't there yet. Chips are still taped out with marginal signal integrity problems despite "simulating ok". Mind you, the analog portions are given a wide berth as I mentioned above, but who knows if they've fully covered this in the CAD tools or in the formulation of the design methodology. Lots of test vehicle chips will be needed.

    Also, integrating passives can be precarious at best. Chips can have elements such as inductors and capacitors, but they're not area efficient at all, and you'll need external passive components anyway. And if you want power regulation for charging functions and battery regulation, fuggeddaboudit. These structures are particularly area inefficient. I don't think that this is what they're trying to do, but if you think we'll have literally everything integrated onto one chip, it won't happen.

    I also have very little faith in the process technologies. If you look at some of the problems that 0.13um manufacturing has had with via voids and low-k dielectric brittleness that have been shown in the trade journals lately, I'd be very nervous with releasing something like this with just anyone's process. TI seems to be better for manufacturability, but TSMC or UMC? Don't count on it - yet. To accommodate the highly integrated nature of this device, they need a small process technology with very rigorous manufacturing capabilities to avoid some of these problems.

    Finally, integrating analog RF and digital requires advanced packaging technologies. If I've got the output to an antenna block in my chip package, how do I get it out? Most likely, this would go into a flip-chip package to accommodate the high integrated nature of this. Well, the flip-chip redistribution layer, the package substrate, and the surrounding pins will all have to be very carefully designed so that the RF signal will be sufficiently isolated. On RF-only chips, this isn't a problem. Heck, they have fully-integrated Bluetooth chips. But Bluetooth only has enough power to reach 30 meters. We're talking a signal that has to reach several kilometers here. That's a difference. It's doable, but it is just another big constraint on the design.

    Can they do it? I think they *might* be able to, but not without significant design effort. Personally, I think they're better off going with a multi-die package and leaving the RF block as a die right beside the other, and specially route through the substrate with its own power. Integrated doesn't always have to mean "everything on one chip". Just like gift wrapping multiple presents in the same wrapper, I think this would be a better way to go for this effort, and will deliver fruit MUCH faster than what I believe they're implying in the article.

  5. Bluetooth has already done this by ceranta · · Score: 2, Informative

    Most Bluetooth vendors have already developped and are in production with 'single' chip designs that incorporate both the digital baseband with the analog radio and all the 'glue logic' in between. This isn't really news on the Analog-Digital single chip designs, but more for the Analog-digital cellular single chip designs.

    Cellular chipsets require very precise parts and separate the analog from the digital for good reasons - noise, crosstalk, coupling, etc. This is a good step forward for wireless design as a whole.

  6. OMAP Comparison by tiomapengineer · · Score: 2, Informative

    This will NOT be a competitor to OMAP. OMAP is a chip that contains an ARM925T RISC, C55x DSP, and just about every peripheral you can think of (USB, MMC, Memory Stick, UART, Bluetooth, etc).

    TI is planning on producing a chip that combines into a single chip the software, baseband technology, applications processing, power management, radio frequency and embedded memory that typically require separate processors.

  7. NYT Article by asv108 · · Score: 3, Informative

    There is also a NYT Article.

  8. The radical part is... by JGski · · Score: 2, Informative
    ...market leader Intel adopts SiGe *heterojunction* technology. For literally decades GaAs folks have crowed about how GaAs will be the technology of the future that would wipe out Si. GaAs has been the traditional bastion of HBTs. The adoption of SiGe now potentially turns that claim on its head, certainly keeping GaAs marginalized as always. Conventional Si technology and economics are still completely available to such HBT designs, which GaAs has lacked - key being having a native insulating oxide.

    What's significant about SiGe and heterojunctions is that current Si technology is homojunction with a fixed, indirect bandgap (the latter being why there are no Si electro-optic devices like LEDs. Heterojunctions allow you to tune the bandgap and even create direct gap devices (which LED/Laser consistuents GaAs, GaInP, GaP, AlGaAs, et al., are) out of indirect gap elements. This throws in an additional set of parameters into the circuit design mix that allows traditional limits on carrier mobility, intrinsic carrier concentrations and other basic device parameters to be thrown out the window. This completely changes both the upper bounds of performance and potentially even basic device operating modes. Many of the "tricks" from the GaAs world become available to "mere mortals of the commercial Si world" such as HBTs, HEMTs, LEDs, EOs, et al.

    Now one of the largest Si manufacturers has seen the economics as workable for general purpose uses. That is profound because for >30 years, GaAs has never gotten there beyond its very small niches, largely due to economics.

    As mentioned, mixed signal devices have been around for some time (every cellphone has a mixed signal IC). Combining digital computing with analog circuitry has often required trading performance on one or the other - often what makes good digital gates MOS devices and processing isn't optimal for analog circuits which is best done in bipolar. HBTs are a special high-performance bipolar technology - an analog designer's dream, yet all the VLSI digital can be on-chip without compromise!

    The TI OMAP comparison is completely out in left field as others have mentioned. Irrelevant.

    JSki