This invention relates to a field-programmable analogue processor.

There are many situations in which analogue signal processing can offer significant advantages over digital signal processing. For example, digital processing of signals conventionally requires the steps of sampling a signal, analogue-to-digital conversion, digital processing, and digital-to-analogue conversion. In contrast, analogue signal processing allows a signal to be processed directly, thus providing a considerable saving both in the number of components required to perform a processing operation and a reduction in the time required for the operation.

Analogue signal processing is currently under-utilised. One reason for the relatively low usage of analogue processing is that the design of analogue circuits, which commonly is at component level, is a complex process requiring a considerable investment of time in both design and testing of a circuit. Analogue design is very specialised, requiring a detailed knowledge of the response of individual components, and consequently is expensive. Thus, when faced with a signal processing task, a designer may be willing to sacrifice the speed and simplicity of a custom built analogue circuit in favour of a less efficient but more easily realised digital circuit.

An analogue processor which could be configured via a computer interface to perform required tasks would clearly be an extremely useful and beneficial development. Such a processor would allow a relatively inexperienced person to produce a required analogue circuit. Furthermore, the processor could be configured to perform a given task very rapidly, a considerable advantage over the current standard development process which comprises circuit design, manufacture, prototype testing and redesign.

One known design of analogue signal processor uses switched capacitor circuits based on digital technology. Conventional operational amplifiers within the processor are provided with feedback loops containing switchable capacitors which are switched on or off to select required mathematical functions. Whilst this approach

provides an adequate programmable analogue processor, it is rather bulk) and contains a large numbers of components which occupy a large area.

The concept of a different computational approach to VLSI analogue design was described by D. Grundy in a paper published in 1994 (Journal of VLSI Signal Processing 53, 8 (1994)). The paper describes how a programmable analogue processor could be realised by reducing a required process to a series of fundamental mathematical steps, these steps being ADD, NEGATE. LOG, ANTILOG. AMPLIFY. (EXPONENTIAL) and DIFFERENTIATE. The paper suggest that any process capable of being broken down into a series of mathematical steps could be programmed into an analogue processor by providing the processor with cells each of which can be programmed to perform any one of the steps, selecting the steps to be performed by individual cells, and connecting them in the required order.

The configuration of programmable analogue circuit suggested in the paper comprises two strings of cells connected in series, each string receiving an input signal. Each cell is programmed to periorm a mathematical function chosen from the above list, and the output from one cell becomes the input to a following cell in a series of cells. The two strings may be linked together to perform functions (for example division) which require two inputs and a combination of cells. An experimental circuit which has been used to generate a logarithmic function is illustrated in Figure 7 of the paper. The circuit uses silicon junctions to provide logarithmic functions and resistors to convert voltages into current. The circuit has the advantage over switched capacitor circuits of more functions, wider dynamic range, real time operation and speed for a given device size. A disadvantage of this design of circuit is the requirement for external components to set gain and RC time constraints. The advantages outweigh the disadvantages however in many applications, and the present invention is related to a practical device for implementing a system of the same general type as that described in the above paper.

It is an object of the present invention to provide an improved field programmable analogue processor.

According to the present invention, there is provided a programmable analogue device comprising an array of cells each of which is controllable to perform any one of a predetermined set of analogue functions, and means for selectively interconnecting the cells to define an analogue circuit between a device input and a device output, wherein each cell comprises an array of subcells each of which is designed to perform a respective one of the predetermined set of analogue functions, each cell comprises an input bias circuit, each subcell comprises a series switch which may be selectively switched to a conductive state so as to connect the subcell to the input bias circuit, each cell comprises a function control circuit which selectively switches on one of the series switches in dependence upon a function select input, each subcell comprises a differential pair of transistors which when electrically coupled by the respective series switch to the bias circuit define an operational amplifier with the input bias circuit, and each cell comprises an output circuit connected to each of the subcells such that the output circuit delivers an analogue output dependent upon the function of the operational amplifier defined by the input bias circuit and the differential transistor pair to which the input bias circuit is connected by the conductive series switch.

Preferably, further switches are provided each of which forms part of a respective subcell and is controlled by the function control circuit to be rendered non- conductive only when the series switch of the respective subcell is conductive, the further switches being arranged when conductive to minimise the effect of the associated subcell on the operation of the device.

The further switches may be connected to shunt resistive components of the associated subcells.

The differential pair of transistors of at least one subcell may be connected to an associated resistive component by an isolating switch which is connected in series between the resistive component and the differential pair of transistors, the isolating switch being rendered conductive only when the series switch of the subcell is rendered conductive. In circuits in which the differential pair of transistors is coupled to source and feedback resistors, respective isolating switches may be connected in

series with those resistors. Λ shunt switch may be connected between the two isolating switches.

Embodiments of the present invention will now be described, by way of example, with reference to the accompanying drawings, from which, for ease of explanation, voltage sealing has been omitted, and in which :-

Fig. 1 is a schematic representation of a multiplication operation which may be performed using standard circuits in an embodiment of the present invention;

Figs. 2 and 3 are schematic representations of different mathematical operations which may be performed using standard circuits in accordance with the present invention;

Fig. 4 is a schematic representation of a cell structure in a programmable analogue device in accordance with the present invention set up to perform the mathematical operation represented by Fig. 1 ;

Fig. 5 is a circuit diagram representing one of the twenty cells shown in Fig. 4;

Figs. 6, 7, 8, 9, 10. 1 1 , 12 and 13 show respective subcells incorporated in the circuit illustrated in Fig. 5;

Fig. 14 is a schematic representation of a three-dimensional structure defined by stacking the subcells of Figs. 6 to 13 and interconnecting the cells in a way which minimises impedances between cells; and

Fig. 15 is a diagram of a circuit which is an alternative to that shown in Fig. 5.

Referring to Fig. 1 , the mathematical operation of multiplication of two operands A and B is represented. Each of the operands is input to a respective cell 1 and 2 which converts the analogue value appearing at its input to an analogue output representing the logarithm of that value. A cell 3 then adds the two logarithms and a cell 4 converts the resultant logarithm to an analogue value representing the product of the analogue values represented by the operands A and B.

Fig. 2 represents the mathematical operation ol ^' division, the cell 5 negating the analogue value at its input.

Fig. 3 represents the operation of raising the operand A to the power B. It will be appreciated from Figs. 1 , 2 and 3 that many simple and more complex

mathematical operations can be performed by an appropriate network of functional cells each of which is in itself a relatively simple circuit.

Fig. 4 represents a device in accordance with the present invention incorporating twenty cells arranged in two rows of ten cells each. Adjacent cells in each row are directly interconnected and the output of any one cell in each row may be connected to a second input of a respective cell in the other row by appropriate switching devices (not shown). Each cell may be switched to any one oi ^' eight conditions, that is non-inverting pass (NIP) in which the cell operates as a unity gain buffer stage, add in which the cell operates to add signals applied to two inputs, negate in which the cell changes the sign of the input signal, log in which the cell produces an output representing the logarithm of the input, alog in which the cell produces an output representing the anti-logarithm of its input, rectify (RECT) in which the cell produces an output corresponding to the rectification of the input, and auxiliary (AUX) which facilitates the connection of external components to perform extra functions such as integrate, differentiate or the like. Each cell can also be switched to an off condition in which the cell is dormant. The cells shown in Fig. 4 have been switched to a configuration in which they perform the function represented by Fig. 1. It will be noted that in this simple configuration many of the cells are acting merely as buffers or are dormant. The cells could be configured, however, to perform complex analogue functions.

Fig. 5 is a circuit diagram illustrating the circuit of one of the twenty cells shown in Fig. 4. Input signals representing operands are applied to input terminals 6 and 7. When the cell is to operate in addition mode, the two signals to be added would be applied to inputs 6 and 7. In all other modes the input signals to be operated upon would be applied to input 6. The cell output appears at output terminal 8. The cell is powered from terminals 9 and 10 which carry respectively plus 2.5 volts and minus 2.5 volts. The cell is controlled by a digital data input applied to terminal 1 1 and a clock signal applied to terminal 12.

The cell of Fig. 5 comprises seven subcells and an output circuit, these eight circuits being represented, respectively, in Figs. 6 to 13. Each of these circuits is

controlled by the output of a respective nand gate 13 and associated inverter 14, the nand gates 13 being switched such that the output of each of them is high only when the three-bit output of an array of three fiip-fiops corresponds to a respective one of the eight possible values for such a three-bit output. The binary values arc represented in Fig. 5 by 100, 110, Oi l , etc. Thus each of the subcells can be controlled by the application of appropriate digital control signals to terminal 1 1.

The output circuit of Fig. 13 comprises an input bias circuit defined primarily by transistors 15 and 16, and an output stage defined primarily by transistor 17 to 23. Each of the subcells of Figs. 6 to 1 comprises a differential pair of transistors 24, 25 connected by series transistor switches 26 to a line 27 coupled to the collector of the transistor 15. Each of the transistor pairs 24 and 25 is also connected by lines 28, 29 to transistors 17, 18 and 19.

The switches 26 are controlled by the outputs of respective inverters 14. Thus the switches 26 are normally off with the exception of the one switch associated with the nand gate 13 selected by the output of the digital control circuit.

When a switch 26 is rendered conductive, current flows from the input bias circuit 15, 16 and this in effect completes the circuit of an operational amplifier incorporating the respective pair of transistors 24. 25. The output appearing at terminal 8 is thus a function of the signals applied to input terminals 6 and 7, the function being defined by the circuit associated with the active subcell as selected by the conductive series switch 26. The circuits represented in Figs. 6 to 12 respectively perform the functions of non-inverting pass. log. add, negate, anti-log, auxiliary and rectify.

Depending on the detailed structure of the circuit components associated with the transistor pairs of each cell, the provision of the series switch 26 may be sufficient when that switch is rendered non-conductive to prevent the existence of the circuit associated with that switch from significantly affecting the performance of the circuit as a whole. Where a subcell incorporates resistive components, however, it is desirable to provide auxiliary switches to minimise the shunt effect of those resistors. For example, the log function subcell (Fig. 7) incorporates a shunt switch 30

controlled by the output of the respective nand gate 13. Shunt switches 30 are also provided in the add function subcell (Fig. 8). the negate subcell (Fig. 9), the alog subcell (Fig. 10) and the rectify subcell (Fig. 12).

In the cell structure represented in Fig. 4, it is desirable to minimise resistive path losses between adjacent cells. This is particularly important to maintaining the accuracy of operation of log and anti-log functions. In general, for every millivolt of attenuation existing between log and anti-log functions, there is an approximately 4% gain error, regardless of the input signal level to the log stage. Therefore, whereas in general electronic applications several millivolts of attenuation between successive circuit stages may be tolerable, this is not so in circuits as described with reference to Figs. 4 and 5.

In order to minimise resistive losses between adjacent cells, it is desirable to fabricate the circuit represented by Figs. 4 and 5 by stacking cells (one of which is shown in Fig. 5) one above the other as represented in Fig. 14. Referring to that Figure, the nth cell 31 has an input terminal 32 extending across its full width and an output terminal 33, also extending across its full width. The nth cell is stacked immediately above the (n+l )th cell 34 which has input terminal 35 and output terminal 36. The conductive tracks 33 and 35 are connected together, minimising cell to cell attenuation given the large width of the tracks and the short length of the tracks. With such a structure the overall resistive path can be limited to the order of milliohms. As a result, attenuation losses are negligible.

In the circuit described above, cell function selection is achieved using the series switches 26 to control current into the associated differential pairs 24. 25 and the shunt switches 32 to shunt resistive components associated with the differential pairs. To select a particular function, the subcell responsible for that function is enabled by rendering the shunt switch 30 non-conductive to thereby release its associated resistor and rendering the series switch 26 conductive. This arrangement works well in terms of isolating the unused subcells, but there is a disadvantage in that current shunted to ground through the shunt switches 30 of the subcells which are not in use represents an unwanted use of power. This disadvantage can be overcome by

introducing an isolation switch in series with the subcell resistors. The introduction of such an additional switch might be expected to produce an error in the subcell function due to the resistance of that switch but this can be compensated, for example, by providing two additional switches, one in series with the source resistance of the subcell and one in series with the feedback resistor of the subcell. Such an alternative circuit is illustrated in Fig. 15.

Referring to Fig. 15, the illustrated circuit components define log. add, negate, alog and rectify subcells. Each of these subcells incorporates a differential pair of transistors 24, 25 and a series transistor 26. Additional isolating switches 37 and 38 are provided, the isolating switches being rendered conductive only when the series switch 26 of the associated subcell is rendered conductive.

For the logarithmic functions, the resistances of the isolating switches 37, 38 are not directly compensated given the illustrated circuit, but these resistances cancel when the log and alog functions are combined.

In some circumstances the isolating switches 38, 39 can cause problems due to capacitive feed through. This is avoided in the circuit of Fig. 1 5 by the provision of shunt switches 39 which are connected between the two isolating switches.

The described circuits exhibit a number of key features and benefits as compared with existing devices. These can be summarised as follows:-

(a) An analogue chip can be developed very quickly. Software has been developed which can simulate single page designs with high resolution in less than 20 seconds on a simple PC. This means that if necessary dozens of iterations can be run without significant delay. The design enables the software to operate on the basis of a one to one correspondence between the software simulator and the chip itself. Downloading of designs from the PC running the software to the chip requires only sixty bits of information.

Viewing of chip activity may be simple, straightforward and therefore fast since every input/output is brought to a terminal pin.

(b) The device has been fabricated using B1CMOS silicon technology which allows the analogue content to be designed with no compromise using bipolar

components whilst the use of CMOS for the digital components ensures that there are similarly no comprises there. To the user this means that the amplifiers have very low offset and its associated drift, low noise, excellent high frequency performance with bandwidths of 4mhz, and the ability to implement a wealth of proven analogue design techniques accumulated over many years.

(c) The device can be used in many applications, unlike competing devices which are limited to selected sectors such as controllers or data acquisition. The device can be likened to its digital counterpart the microprocessor in that it can be applied to any analogue situation. To the user this means that once an investment has been made in understanding and learning to use the device, this investment does not have to be repeated when changing applications.

(d) The inbuilt structural features which the device brings to the design process means that designs are more predictable and perform better. This structure also means that the design process can be opened up to a wider design community including digital designers.