Personal dimension

About working in big science

DLNP engineer Alexandr Lapkin first learnt about JINR in 2014, when he was a fourth-year student at Saratov State University. His friend Petr Smolyansky that had left for Dubna earlier offered him a job. At JINR, he started working at DCB DLNP, under the supervision of Alexey Zhemchugov and Georgy Shelkov. Alexandr wrote his bachelor's thesis at JINR and later, entered the master's programme at MIPT. On the pages of our newspaper, he talks about his work and it seems that for the first time, we are publishing details about such a complex topic as electronics for nuclear physics.

The first impression of Dubna and JINR was a shock. The beginning of an independent life, far from family, the first job. And work in big science, with big scientists. The first tasks were more complex than educational ones, but they were lost against the background of the department's activities. At first, it was overwhelming, but over time, you gain experience, establish contact with your comrades and get involved in the work.

I work at the Department of Colliding Beams in G.A.Shelkov's Group. We develop a multi-energy X-ray tomograph for medicine, geology, archeology and other applications. Multi-energy tomography differs from conventional tomography in that it is not the total absorption of radiation by the object that is saved, but the dependence of this absorption on the radiation energy. The dependence of the absorption index on the radiation energy is different for each element and has its own characteristics. Knowing these dependencies, it is possible to determine the spatial distribution of various elements in the object. The object can be, for example, a mouse with various radiopaque substances injected or a rock sample which it is necessary to determine the content of useful minerals in. But it requires a detector that can not only determine the radiation intensity at a point (pixel), but also determine the radiation intensity at a given energy. Or, in particular, the detector should measure the energy of an individual photon. Semiconductor pixel detectors are suitable as such detectors.

In the Group, I am engaged in semiconductor detectors, electronics and software for them. In particular, I develop readout systems for semiconductor pixel detectors. The basic part of such systems is FPGA, in working with which I have gained some experience.

FPGA (Field Programmable Gate Array), or as it is also called, PLD (Programmable Logic Device) consists of many different units, such as memory units, adders, multipliers, dividers, standard units, phase-locked loop (PLL) units, high-speed interfaces or even a processor. Standard units that are the majority in PLDs are defined by manufacturers and often consist of registers, look-up tables, multiplexers, adders... PLD units are connected by programmable connections. This architecture allows PLDs to execute any user-defined function if there are enough resources to implement it, PLDs are programmed using special programmes in HDL language and recently, in a high-level programming language (C++, Python).

Development and debugging of such devices is longer, more complex and more expensive than conventional software, but cheaper, simpler and faster than conventional microcircuits. At the same time, PLDs operate faster than processors, since data processing is not at the programme level, but at the hardware level. The processor has a constant architecture that implements a universal set of simple operations - processor instructions. The sequence of these operations is the computer programme. On the other hand, the PLD architecture can be programmed to meet the task into a best-of-breed solution. It is not the technique of meeting the task that is adjusted to the architecture, but the architecture to the task. Thus, PLDs provide great opportunities for parallelization and pipelining, that is, for simultaneous execution of a task at different stages with different chunks of data. In addition, unlike microcircuits, they can be reprogrammed. Due to these characteristics, PLDs are ideal for developing experimental and small-scale devices for converting data from one interface (detector) to another (computer). During our work, our Group has developed and tested several reading systems for pixel detectors.

The next idea after developing the reading system is to transfer part of the data post-processing from the computer to PLD. It will reduce the amount of data transfer between the computer and the reading system, and speed up the processing, since it starts immediately upon data occurrence in PLD. The processed data takes up less memory.

One of the major stages of data processing from pixel detectors is clustering - the process of combining pixels from one event. A semiconductor sensor is a reverse biased p-n junction. When a charged particle passes through a semiconductor sensor or when it absorbs a photon, a cloud of free charge carriers is generated that moves towards the pixel electrodes under the impact of an electric field. But since equally charged carriers move in one direction, the carrier cloud becomes larger under the impact of diffusion and electromagnetic forces. Thus, several pixels are triggered by one event or particle. To more accurately determine the coordinates of the flight or absorption of a particle, it is necessary to specify which pixels belong to one event. In addition, if the detector is able to measure the particle energy by the charge accumulated in the sensor, then it is necessary to add up the charge for all the pixels of one event.

Part of my work is developing a clustering algorithm for PLDs. It has its own characteristics: it has to operate at data occurrence and not all data is available at once. The order of pixels in the output stream can generally be any. That is, it is possible that two clusters are obtained based on the accumulated data and the next pixel unites them into one. At the moment, an algorithm for sequential clustering has been developed. It is based on sequentially enumerating pairs of clusters among the already gained clusters. To significantly reduce the operating time, sorting of the incoming pixel stream by coordinates is used. It allows us, on the one hand, to reduce the number of comparisons that increases the operating speed and on the other hand, to reduce the memory required for operation. This algorithm has been tested on current reading systems and on randomly generated data. However, it does not save data on the cluster shape, only its general characteristics: coordinates, total amplitude, time... A new algorithm is currently developed that will save the shape of clusters. This algorithm is scheduled to be used for new semiconductor pixel detectors, as well as in the SPD detector.

Tomography in our Group is carried out using several facilities, including the experimental tomograph "Kalan". I participate in the development of software for controlling and configuring the tomograph, in particular, for controlling the X-ray tube. I also participate in data processing and configuring the Timepix/Medipix detectors and in development of an algorithm for identifying items. I have developed software for processing data from the Widepix detector and software for controlling the Galapad detector.

In addition, our Group works with new semiconductor detectors. In particular, we have developed software for the Timepix 4 detector. The software and the detector itself have been tested on the SPS and PS test beams at CERN together with the Straw Tracker RD Group that develops a straw detector for the SPD. Data analysis is currently underway.