Products planning 


From simple to complex, AMT plans to follow the development roadmap (as shown in Table-1) and focus on the three product lines in parallel, i.e. low & medium density product line, high density product line and neuromorphic product line.

 

Table-1: AMT’s Product Planning


Product category

2018/4Q

Standalone Low & medium density PCM products- EEPROM & NOR

 

2019/4Q


Embedded low & medium density PCM products and special PCM products(FPGA/PLD,TCAM, MCU)

2019/4Q

(2D Xpoint)High density PCM products phase Ⅰ

2020

Reconfigurable computing memory

2021

High density PCM products phase Ⅱ (MLC, 3DXpoint) and neuromorphic memory products

 


The following is the product introduction:

Table-2: AMT’s Future Memory Product Introduction

Product category

Product introduction

EEPROM (Electrically Erasable Programmable Read Only Memory)


The typical advantages of EEPROM are non-volatile, erasable and fast programming. It’s not necessary to move program out of EEPROM system such that it is more convenient, more effective, more feasible to store, refresh and program data. EEPROM can be categorized into serial one and parallel one.

1) Serial EEPROM- SPI and IIC types are the most popular ones. They can be applied to all kinds of flexible memory devices to store personal preference data, system calibration and configuration data, and data with high written frequency.

2) Parallel EEPROM- Byte-alterable parallel access memory device with excellent data retention, high quality and manufacturability. It is mainly applied to direct code execution and high reliable data storage applications, such as telecommunication, air electronic equipment and military field.

Initially, AMT will utilize 45nm process to manufacture products of 1M EEPROM with SPI/DPI/QPI interface. After that AMT is going to produce SPI EEPROMs (from 512k to 4M) and IIC EEPROMs (from 512k to 8M).


NOR Flash


NOR flash memory can be used as execute in place (XIP) memory. It has higher efficiency and faster read. It mainly can be used to store programs with the highest cost-performance gain from size 1MB to size 4MB.  

There are two major types of NOR flash, serial and parallel type. Due to shortcomings of low integration and complicated device structure with multiple pins, parallel NOR has been replaced by serial NOR which has high integration and simple device structure with only few pins.


Embedded PCM products of low & medium and special PCM products-

PCRAM,FPGA/PLDTCAM, MCU


- FPGA (Field-Programmable Gate Array). FPGA is suitable for diverse markets, such as telecommunication market, the field of industry control, auto, aerospace, consumer electronics (especially TV), etc. AMT’PCM-based FPGA can save 70% in chip area and power consumption compare to traditional FPGA. AMT plans to enter the 55nm/45nm FPGA market with low & medium gate counts.

- TCAM (Ternary Content-Addressable Memory). TCAM is mainly used for high-speed network router. PCM-based TCAM consists of two transistors and two PCM resistors in each unit. Hence its chip area and power consumption are a quarter of traditional TCAM, which is very expensive.

- MCU (Micro Controller Unit). Embedded memory in MCU can have different applications in different scenarios, for example:

1) ROM/EEPROM is used as non-volatile Bootloader memory.

2) RAM is used as on-chip cache to store running program data

3) Flash(NAND/NOR) is to store data and program

MCU is mainly used in the application fields of consumer electronics, computer, auto electronic and industry control. Auto electronics, industry control and IOT/wearable fields are faster growth MCU markets.


Re-configurable Computing memory


Reconfigurable computing is a research project which was awarded a National Second Prize by Microelectronics Institute of Tsinghua University in China. The project is also supported by Intel. Reconfigurable computing aims at achieving high efficient dynamic reconfigurable computing through hardware & software co-programming. The technology is a breakthrough compared with the traditional model in the sense that it can produce a reconfigurable computing chip based on the basic logic of “circuit follows algorithm change, structure follows application change”. The main idea is to build a connection network of internal parallel processors. In order to achieve a substantial increase in the performance of processor, the connection network can be reconstructed according to needs of data. If a processor of configurable computing uses a nonvolatile memory, it can improve energy efficiency and then solve the problem of huge power consumption in Internet Data Centers (IDCs). AMT has signed a cooperation agreement with Microelectronics Institute of Tsinghua University.

Shutdown management of PE and memory can be achieved by using PCM as part of PE’s memory. When the PE is restored to work, the data and program will not be lost and the operation of data can be maintained. Hence it can largely reduce system energy consumption and promote energy efficiency. 


High density PCM products phase Ⅰ


 - 2DXpoint / PCRAM is based on phase change memory. It’s a nonvolatile random memory. It can replace DRAM because of advantages of low cost, low power consumption and high reliability in future.

High density PCM products phase


Initially, AMT plans to focus on different architecture research of MLC, Cross Point and Process Macro, gradually accomplishing process technology and products functional verification of high density PCM (128Gb/256Gb) beyond 20nm process node.

With accumulated market knowledge and AMT’s production experience of PCM, AMT will enter the field of high density memory (128Gb/256Gb) such as SCM, U disk and SSD with domestic IDM together.

Neuromorphic Memory


PCM-based neuromorphic chip has many advantages such as good fatigue resistance, good data retention, fast erasure, nondestructive read operation, good compatibility with CMOS process. Hence it can achieve functions (e.g. study and memory, etc.) similar to human’s neuromorphic memory network. Neuromorphic chip will be more suitable for multiple applications, such as large data analysis, multi-function sensors (electronic nose), early induction networks, “smart” personal assistant, guide glasses for the blind, self-driving car, auto robots, and so on.