Under The Hood: Semiconductors In Audio

Semiconductors are essential building blocks of most modern electronics, with the progression from discrete transistors to integrated circuits largely universal, chips dominating usage. To get an overview of the practical implications of trends in the semiconductor market, Pro Sound News periodically queries audio design engineers about their semiconductor usage.

Apogee’s Symphony I/O is an example of a product package made possible by the use of surface mount components.

Image placeholder title

Semiconductors are essential building blocks of most modern electronics, with the progression from discrete transistors to integrated circuits largely universal, chips dominating usage. To get an overview of the practical implications of trends in the semiconductor market, Pro Sound News periodically queries audio design engineers about their semiconductor usage.

Participating in this year’s survey were Great River Electronics Project Engineer Dan Kennedy; Senior Design Engineer at Apogee Digital, Lucas van der Mee; B.J. Buchalter, VP R&D at Metric Halo; Crane Song President Dave Hill and from Lynx Studio Technology Co-founder and Chief Hardware Engineer Bob Bauman.

Four of our virtual panelists agree that ICs they have traditionally used are harder to source, with van der Mee calling it “a common problem nowadays.” Parts obsolescence has forced Apogee to discontinue some products. “Some of our older products we cannot repair,” he shares, “simply because parts are not available anymore.” Only Bauman says his company was not having such problems, adding “I hope it continues this way!”

Great River has addressed the problem in advance with lifetime buys of critical chips, says Kennedy, while Buchalter says Metric Halo primarily redesigns circuits as needed. It’s a mix of the two approaches for Hill, who says that some circuits require redesign and in other cases it’s “buy as you can now and hope for the best,” as substitutions don’t always exist.
“With new designs,” says van der Mee, “we always check with manufacturers about the life of a part. It can be even as simple as asking which footprint will have the longest life as sometimes just a certain unpopular size/configuration of the part is being discontinued. We have redesigned a few times as well in order to keep a product on the market.” For repairs of older products where components are no longer available, he adds that “I sometimes design a “dead bug” circuit or mini PCB to replace an obsolete part with something that can replace its function.”

Hill and Kennedy report that their use of surface mount devices (SMDs) is increasing. These tiny, robotically placed parts don’t have the ‘legs’ of chips of old and are closer to the size needed by the circuitry internal to the chip rather than the size being dictated by the traditional packaging and the need for leads that can extend through printed circuit board holes. This makes field servicing difficult at the component level, but is necessary to achieve the PCB density needed for more compact circuits. SMD parts have all but replaced through-hole design chips. Baumann says that Lynx has “always used SMD parts extensively” while van der Mee reports at Apogee, “We do surface mount only and have been doing so for decades. The only through-hole parts are connectors, switches, encoders—that kind of stuff.” That’s the same for Metric Halo, Buchalter reports, adding bulk caps, transformers and pots to the list. “Our designs are primarily surface mount and always have been,” he explains. “If we can find a cost effective SMT [Surface Mount Technology] component, we use it as they are easier to assemble and easier to service.”

Our respondents largely do not fall back on favored legacy components. “It is way too risky (obsolescence) and new parts are simply too good to ignore,” offers van der Mee. “Parts have become so much better over time. I also enjoy the challenge of designing with new devices. New options can lead to new ideas.” The exception in our replies was Bauman’s citing of “older Cirrus converters that we still use because they continue to outperform newer devices.”

Integrated circuits do not operate in isolation. Discrete semiconductors (transistors and diodes) and components like capacitors and resistors are necessary in electronic designs. Kennedy reports that film caps and FET's” are tougher to find.” Hill adds conventional transistors to the “big problem” list, saying that “some thru hole transistors are not longer made and there are no SMD versions.” Bauman cites long lead times for small signal relays (“longer than any other component). Metric Halo’s problems with sourcing peripheral components haven’t been too extensive, according to Buchalter. “An occasional diode has gone obsolete, and there is always a good replacement,” he elaborates. “We did encounter some problems with the transition to RoHS where component performance changed [the Restriction of Hazardous Substances laws that went into effect a few years back forced changes in electronic component manufacturing]. But that has long been sorted out.”

van der Mee also laments the rapidly declining availability of transistors and jFETs, “I guess as a result of the shrinking market. It is a shame, as I still like to do some discrete design here and there. The rest of the components I usually have no problems with. They are pretty universal, so lots of crosses are available. But it does happen. We recently had to do a quick spin on one of the Quartet PCB’s, simply because the common mode chokes (for EMI) were obsoleted and there was no cross anymore in the same size.

Our respondents have migrated to new opamps (operational amplifiers, standard building blocks in audio circuits) over the years, with TI/National components getting all the specific kudos (TI purchased National Semiconductor a number of years back), like Hill’s complements for the OPA16xx series. New demands for portable and battery/USB powered devices puts new demands on components. “In the last few years I have had to design more and more low power circuits as a result of the strict power budgets of iOS devices and battery operated units,” van der Mee explains. “So finding good audio parts is not easy.” He lists TI’s LME49726 and THS4531 as examples of chips that have given good performance.” Buchalter says he also has found “newer opamps that provide exceptional performance with very low power requirements.”

Hill says that he hasn’t adopted any new components lately, nor Kennedy, except “incrementally improved ones.” Hill says he’s using components “that are not easy to use.” Buchalter says he’s adopted new chips in an “evolutionary way,” while van der Mee says “I am always looking for new stuff and do use new components all the time.”

Analog component improvements have come primarily “in performance vs. price and power,” says Buchalter, which “allows for denser designs.” van der Mee offers two examples of where he’s seeing improvements in analog components: “Fully differential opamps have become pretty common in my designs in the last five years. They are wonderful parts as they combine into one circuit I used to need three opamps for.

“The other one is analog switches. I have learned to implement these with almost zero losses. Relays are now only used in configurations where the use of an analog switch would compromise the audio quality. The advantage is cost and size. It allowed me, for instance, to design the insert routing matrix for the Symphony IO preamp. It has 256 analog switches (and 64 opamps) on a circuit roughly the size of a playing card. Performance is stunning.

“In general you can say that the quality of (cheaper) parts has gone up a lot. You just get more bang for the buck. By using better components I am able to keep the circuits simpler, resulting in a much cleaner signal path.”

Mic pre switchers are the only modular analog parts (dynamics engines, balanced drivers and receivers, VCAs and other circuits built into a single chip with minimal external components) that van der Mee says he employs. “THAT’s 5171 is at the heart of most of my mic pre design gain switching, he elaborates. “It is a great part as it allows for your own implementation/configuration.” The THAT 5171 “allows digital gain control of an analog signal and it works well,” echoes Hill.

Kennedy also looks to THAT for line receivers and drivers (balanced input and output circuits on a chip), and also uses “PGA's from TI [Programmable Gain Amplifiers]. Otherwise, generally, dual opamps and CMOS switches make up the bulk of the IC's in our audio products.” Bauman says that Lynx uses “a mix of balanced receivers and discrete designs for analog inputs.” The receivers are used “in cases where space is an issue.”

Digital semiconductor devices include converters and DSP components. The launch of new professional quality components has slowed, with research no longer focusing on improving the quality of digital conversion devices as much as on low power operation. Despite that, van der Mee notes that “there are still interesting developments. It is nice to see AKM work very hard to get true high-end chips in the market. ESS is of course another powerhouse for high-end conversion. We were the first to use their DAC’s in the pro audio market.” Hill is seeing some interesting components being introduced; Bauman agrees, noting that “AKM apparently has some new devices coming to market that look promising on paper.” Kennedy uses no digital components, save for microprocessors for control.

Hill calls out AKM’s AK4490 specifically as a DAC chip with “better filters and linearity.” Cirrus Logic parts had “been my choice for a decade,” says van der Mee, but that Cirrus “seemed to have lost interest in our side of the market.” He says he’s employed an AKM chip into an ADC for the first time—the AKM AK5388, a “good, affordable, solid performer.” He adds that chip maker ESS “pulled it off again” with its ES9018K2M Reference DAC, describing the DSD/PCM, 2-channel DAC with integral volume control as “even smaller and even more efficient, giving unrivaled performance. The chips are great sounding.” The number of applications is growing for the ESS parts used at Apogee.

For getting digital audio in and out of devices ranging from DAWs to loudspeakers, a broad array of interface standards are deployed, including USB, Ethernet, AES50, FireWire, Thunderbolt, AES3, AES10 (MADI), ADAT optical and the list goes on to now include Ethernet-based protocols. Crane Song is sticking with AES3 and ADAT optical for now, while Lynx has added a Thunderbolt card to its converter options.

At Metric Halo, a FireWire pioneer, Buchalter says that USB, Thunderbolt, MADI and Ethernet are in use. On Ethernet he says, “we have implemented MHLink–which is an exceptionally low-latency (1 sample), high-bandwidth (128 channels/direction/port @ 192 kHz) point-to-point link over GigE. Our use of GigE is soft, however, so we have left open the possibility of implementing one or more of the more generic networking protocols (Ethernet AVB, AES67, Dante).”

For Apogee, Thunderbolt and USB are the interfaces of choice, being “universal and standardized,” says van der Mee. “Ethernet has our attention,” he continues,” but we haven’t included it yet for the lack of a true universal standard for audio (and video) application. We are still supporting AES3, SPDIF and ADAT but they are of limited use nowadays, so we only provide a few connections.” XMOS, Cypress and Atmel provide the chipsets Apogee uses for USB. “Thunderbolt is, of course, an Intel based system, with a few peripheral parts from other manufacturers.” The Intel DSL3510 Thunderbolt controller is deployed by Lynx, and while Bauman says that the devices are adequate for the task, “the certification process is very long.” While Apogee is largely satisfied with the current crop of interface devices, Buchalter says Metric Halo transitioned to digital I/O implemented in FPGA logic (Field Programmable Gate Arrays, flexible devices that can be programmed to perform a broad swath of audio applications), “We have found that the current interface devices are either too expensive, too limited or have obsolescence problems and have decided to decouple ourselves from those issues,” he explains.

Our panel is keeping tabs on movement towards interoperability between Ethernet audio standards (OCA and AES67) save for Hill who explains that “all the new formats are a large task for a one person operation and that is before you need to deal with drivers for PC and Mac that change each time they change the OS,”

For DSP designs, Hill notes that there are some advantages from one family of processors to another, and says that most of the DSP design toolsets “have problems.” Buchalter cites five primary considerations for DSP component selection: “Price/Performance, Available memory size, Memory Bandwidth, Interfacing and High Level language support.” As for design support toolsets, he says that “some processors have enough memory and good enough compilers that coding can be done in a high-level language rather than assembly to yield good performance. That makes a huge difference in how portable DSP algorithms are from one platform to another and as a result how much development can be spread across multiple targets. We have not seen a substantial advantage in terms of the high-level signal chain tools because we don’t feel that they are particularly useful for our target products.”

Given a chance for a final word on topics covered or questions not asked, Bauman notes that at Lynx, “Most of our products are based on FPGA’s. We can do lots of stuff in one chip and can continue to add features in the field via firmware updates.” Buchalter comments that, “Overall, there has been a nice improvement in performance vs. power while pricing, in general has stayed the same or even come down in some cases. This is great because it supports the kinds of high density, high performance products that customers are looking for.”

The final word goes to van der Mee: “The even further miniaturization of parts is great but also a challenging development. An SOIC-8 part is now huge [the early footprint of SMT opamps]... Most of the opamps I currently use are MSOP-8, which is still large compared to the wave of QFN and ball grid array parts popping up. It is wonderful as you can put so much circuitry on small footprint, but for debugging it is a nightmare. You can barely probe as you would short out several pins or even worse you cannot reach the pin at all. We are growing in a habit of making a large design first with the largest size available of a part and debug that. Then once we are happy with the performance, we’ll shrink it for the production revision.

“The number of new power management parts is astounding. They become better and better. Smaller, more efficient. I really enjoy designing the power section nowadays as there is so much choice, allowing for so many configurations. As a result our units have very elaborate power designs. Which is very important. As I like to say: ask an athlete about their secret? The food they eat, will always be part of their answer.”