Analogue electronics: Focusing on the interface

5 min read

Moore's Law tells us that silicon manufacturing improvements give us twice the number of transistors every two years for the same cost. Another way of looking at this is to observe that the price for a silicon wafer never goes up and, for a given wafer price, we have to figure out how to best use twice as many transistors every two years.

However, Ohm's Law states that analogue design rules don't change with reducing geometry and, as supply voltages reduce, signal to noise ratio (SNR) becomes difficult to preserve, never mind improve. Indeed, if signal swing is halved then, in order to preserve SNR, we need to reduce noise by half. This, in turn, requires a fourfold increase in current in the circuit, resulting in a doubling in power consumption. And so we are presented with the paradox that while small geometry processes with reduced signal swings dramatically reduce power consumption and improve performance in digital circuits, they do just the opposite to analogue circuits (conveniently ignoring the speed improvements that do come to analogue). Meanwhile, the doubling in transistor count similarly increases design complexity, demanding changes in the design methodologies and tools used to implement these more complex designs; gone are the days when the lead design knew what every transistor on the chip did. Despite these dramatic changes in the design environment, it requires only a simple examination of the physical world around us to be reminded that the real world is analogue, and more real world interfaces are required now than ever before. Analogue design is not going to disappear anytime soon, but a great deal of the analogue design now being performed is aimed at implementing these real world interfaces, typically analogue to digital or digital to analogue conversion in all its myriad forms. Some reflection on these issues has led analogue designers to realise that, whilst analogue design problems do not fundamentally change, design architectures and styles need to change in order to take advantage of what are actually opportunities presented by these new rules of changing design economics. From this emerges design styles that either change the architecture of the circuit to implement the function in the digital domain directly (and the analogue circuits now become a/d converter and d/a converter interfaces to this signal processing), or else use 'digitally enhanced analogue'. In this approach, an analogue function is implemented in what might otherwise be a relatively poor performing circuit technology, but is then surrounded by digital circuits such as calibration or correction functions that help achieve an overall improved analogue performance. Audio circuits present some good examples of where both these approaches are taken. As touched on previously, analogue SNRs are very sensitive to power consumption, with a 6dB increase in power typically needed for a 3dB improvement in SNR. But, in the digital domain, a 6dB improvement in SNR requires only a single bit to be added to word length. When compared to the 20 to 24 bits typically processed, this is incredibly good value. Furthermore, in the analogue domain, if two analogue stages are cascaded, each having an SNR of perhaps 100dB, then the resulting combination now is a signal path with an SNR of only 97dB. Clearly, moving audio signal processing into the analogue domain as quickly as possible is very desirable. A quick look at the architecture of most audio systems today says that this is the direction being taken, with signals being digitised as quickly as possible on input to the chip, then converted back to analogue only at the very output of the chip. In most complex audio systems, there will be several audio streams on multiple asynchronous clock domains. Mixing these signals requires conversion to a single sample rate domain, a process which was relatively expensive until Moore's Law gave us all these 'free' transistors. First generation music playing mobile phones would typically mix music and voice signals in the analogue domain because it cost less than doing so digitally, using the relatively large geometry digital gates. The result was worsened SNR for the overall signal path, with the additional potential problem of degradation of the analogue signals while they were routed across the circuit board to the analogue mixer. Whilst the analogue solution was acceptable at the time, the implementation of this function in the digital domain is clearly preferable, both in terms of performance and cost. And, of course, the digital advantage only grows wider as technology develops further. DC offsets are a common problem in audio circuits, causing pops and clicks as signals are enabled and disabled. In older analogue signal path implementations, these offsets would require to be removed typically using external ac coupling capacitors, with associated cost in board area and component count. Nowadays, the analogue stages that remain are normally designed to have near zero dc offset in order to remove the need for these extra components. This is often achieved using a dc servo approach, where the output signal is carefully compared to zero in an a/d converter, the residual dc offset then being calculated in a digital lowpass filter, then removed at the input of the analogue stage by using a d/a converter to apply the inverted version of the calculated offset into a compensating input. These 'digital enhanced analogue' solutions must require tens of thousands of transistors, solely to replace a single external capacitor, but such are the economics of advanced silicon manufacturing that this solution is better. This use of digital circuits to enhance analogue is so compelling that it is now normal to refer to 'analogue' circuits when, in fact, these circuits are really 'mixed signal'. As such, the design tools required to implement analogue circuits are now required to support significant digital circuit content. Of course, the traditional SPICE type simulations tools, usually powered by schematic driven user interfaces, are the first tool the analogue designer picks from his toolbox. But these tools are now required to support higher level digital design approaches, typically Verilog or similar, through cosimulation. Once that mixed signal approach is established, a traditional analogue designer might consider that Pandora's Box has been opened and they are now exposed to all the activities associated with current digital design practice, such as code driven design, synthesis, design for test and verification. The adoption of digital design techniques such as these means there is an inevitable increase in transistor count as extra devices are added to support these design practices, usually with the intention of improving test coverage, or design time and reusability. Yet, as when a single capacitor can cost effectively be replaced with tens of thousands of transistors, these extra 'free' digital transistors allow the analogue designer to achieve better results. Clearly, a major part of the analogue designer's task is now translating in and out of the digital domain. Inventing new methods of doing this with ever increasing accuracy and efficiency, whilst taking advantage of, or alternatively, adapting to the limitations of, new process technologies is a full time activity for analogue designers. Different analogue design styles are being adopted to better suit the strengths of these small geometry processes, and are enabling a/d and d/a converter performance levels to be further improved, despite the difficulties presented by the low voltages these processes support. For example, in the past, very good quality integrated capacitors were available on analogue cmos processes, usually built using dual polysilicon layers or dual metal layers. These capacitors matched well and enabled very accurate a/d and d/a converters to be implemented using switched capacitor techniques. However, these capacitors do not scale well with process technologies and often require extra manufacturing steps to build them. Alternative approaches, such as switched current or resistor techniques, scale better as processes shrink and are now a more common design approach in deep submicron analogue design. Better still are techniques that trade off precision in voltage for precision in time, taking advantage of the ever increasing speeds of advance silicon processes. Smaller geometry silicon enables higher clock speeds for digital circuits, and, similarly, analogue circuits can be built with higher bandwidth or switching speeds. This presents analogue designers and, in particular, those with interests in rf circuits, with the opportunity to implement high frequency circuits right up to the antenna connection. Furthermore, increasing analogue circuit speed enables implementation of ever higher speed a/d and d/a convertors, enabling rf designers to evolve their circuit architectures to take similar advantage of moving into the digital domain as soon as possible, just as audio designers have done. In conclusion, analogue designers will always be required, but they will need to evolve their skills to accommodate use of new techniques and new tools. In particular, they need to grasp the opportunity that having thousands of nearly 'free' digital transistors bring to their circuits. In order to do that, they need to understand how to use these digital transistors to best effect and how to codesign these digital circuits into their architectures. The best of these analogue designers are producing circuits that achieve better overall system performance than ever, despite the challenges presented by process technologies that are becoming ever more optimised to implementing digital circuits. Peter Frith is Wolfson Microelectronics' chief technical officer.