Home         Authors   Papers   Year of conference   Themes   Organizations        To MES conference

Homogeneity Lot of Integrated Circuit Inspection based on Radiofrequency Measurement  

Authors
 Semenov A.V.
 Fedorec V.N.
 Starcev V.N.
Date of publication
 2016

Abstract
 Identification of microcircuits at various stages of their life cycle is an important tool of quality control. The design with such identification has been called Design-for-Security. Many approaches in this area based on the introduction of additional mechanisms for identifying the quality of the chips and protect them against copying at the level of IP block [1, 2]. Due to the limited versatility of the proposed method of detection of anomalies in the monitored parameters of the chips, as well as relatively high cost of such inspections we requires to ensure the possibility of determining such anomalies in microcircuits by comparison of simple characteristics set[3,4]. In this paper we carry out a comparative analysis of different methods of identification that enable us to make the decision without detailed modeling of the chip.
The analysis of signal time delays in the predetermined paths is most widely used for anomaly detection in microcircuits [5, 6]. Primarily discusses anomalies related to the degradation processes that cause the change in the threshold voltage of the transistor p-type due to temperature instability voltage reverse bias (NBTI) and hot carrier injection (HCI) [7].
In particular the approach based on time signatures of electric circuits [8] is widely replicated. It includes three stages: selection of the controlled paths of the signal, the delay measurement and identification. It is used as known patterns of faults (path delay fault, time delay fault), and specialized approuch based on different measurement models and signal processing [9-11].
The closest to our approach is one based on the use frequency scan to obtain a signature delay [11]: as a result of stable selection (taking into account technological dispersion) of frequencies to generate the test patterns. This approach allows us to reduce the identification to the comparison of binary signatures convolutions in the Hamming metric.
A serious drawback from the point of view the practical use is the uncertainty of a sufficient number of ways (paths) to the signal to obtain such signatures. In addition to constructing a decision rule and decision itself there is required fairly extensive preparatory phase of the simulation. Given the known weaknesses of the identification rules and taking into consideration the gaps based on the permissible limits of the parameters and the classical method of principal components [12] such an approach may be recommended for use in the specific product taking into account the actual process variability involved with silicon factory.
When making certification testing of circuits [15] to identify potentially unreliable microelectronics components the numerous control methods of individual identification characteristics [4,5] should be used, which used together allow us to speak about the homogeneity/heterogeneity of the presented batch (lot). These methods also applicable to the detection of counterfeit products and allow us to speak about the "authenticity" of the chips [4,7,13]. As a method of uniformity control we propose to use a new approach based on "internal radio portrait" [16].
Internal radio portrait is a set of amplitude-frequency, phase-frequency and time (impulse response) characteristics, obtained on different pins on the device (in particular s-parameters). (This definition is consistent with the presence of "external radio portrait", which is used, for example in our notification, in the analysis of electromagnetic compatibility [17]).
It may be underline that measuring S-parameters is low energy process and the uses of this technique dont disturb the device under investigation [16].
Analysis of the existing design experience [18], and chip quality control [19] have shown that circuits within the lot can vary due to two factors groups of:
− "natural" causes (of process variation, errors, defects and malfunctions, interference effects);
− "artificial" reasons (bad acquisition parties/issues of technical process, counterfeit).
Our method contains the following steps:
A. Definition of the inputs and outputs for measurements
A random selection of the reference device from the lot, measuring its internal radio portrait, which is considered as the reference Rp_gold (and is used as a feature vector with known class label).
A detailed study of the internal structure of the investigated reference device using destructive methods, the proof of identity based on the obtained characteristics, comparing them with a technical description (specification) or with the material standard.
B. Identification
Getting internal radio portraits to other devices from the lot Rp_1,...,Rp_(n-1), where n is the number of devices in the lot.
C. Result comparison
The comparison of the reference radio portrait with radio portraits for other microcircuits of the lot.
We studied in detail the lot of 121 PCs integrated circuits. A comparative analysis of the s-parameter on-a-chip FPGA GAL22V100-15LPM WITH device AgilentE5071B had been made. Radio portrait measurements were held between pins GND and VCC These may be considered as 2-pole measurement. In addition to measurements of radio portrait we made visual optical inspection, special x-ray control, control the current-voltage characteristics, and functional testing. During the measurements, the differences between the individual chips were obtained.
Comparative analysis of radio portraits were conducted on the basis of the 103 units from the lot. From the initial batch the disabled (defect) products identified in the functional control were removed, as well as products with damaged pins and an identification code that is not related to any of the four detected x-ray clusters.
For characteristics s11 and s21 control accuracy was more than 75% (clustering was based on k-means method).
The results of the monitoring reliability are generally consistent with the size of training samples: type 1 20 units, type 2 22 PCs., type 3 58 PCs., type 4 3 PCs. and with the results of the evaluation S22 characteristics as the least informative in such control.
To reduce dimensions (number of measurements), we can recommend the following heuristics.
1) to control radio portrait between the power supply VDDIO and any opposite pin (applied in the experiment measuring circuit between "Vcc" and "ground" are functional only for very simple circuits with small number of pins).
2) to control only opposing pins. The heuristics take into account the properties of the geometric weights of the graph edges measurement.
The analysis of internal radio portrait of microcircuits fnd other electronic products allows us to control the homogeneity of the batch of devices by nondestructive low-energy method and to search for anomalies comparing with the benchmark obtained by other destructive methods.
It is possible to use the inclusion of the integrated circuits in the radio tract in the form of a 2-pole device, which expands the possible block diagram of the measurement.
Given the fact that there is a proliferation of complex multichip 2.5 D and 3D configurations of chips in one package (for example, Xilinx Virtex7T), the sensitivity of the proposed method to the geometric differences will not lose its relevance.
Keywords
 counterfeit detect, quality control, s-parameters
Library reference
 Semenov A.V., Fedorec V.N., Starcev V.N. Homogeneity Lot of Integrated Circuit Inspection based on Radiofrequency Measurement // Problems of Perspective Micro- and Nanoelectronic Systems Development - 2016. Proceedings / edited by A. Stempkovsky, Moscow, IPPM RAS, 2016. Part 3. P. 49-56.
URL of paper
 http://www.mes-conference.ru/data/year2016/pdf/D185.pdf

Copyright 2009-2019 IPPM RAS. All Rights Reserved.

Design of site: IPPM RAS